The AI Read is on a break. The last edition was November 15, 2026. Everything here stays open to read.
The AI Read

Glossary

AI industry terms and jargon.

Find

1

10-Q

The quarterly report a US public company files with the SEC. It is less detailed than the annual 10-K but often discloses figures, such as customer concentration, that never make the earnings release. Read more

10b5-1 plan

A trading plan a company insider files in advance, setting when and how many shares will be sold, so the sales cannot be accused of using inside information. Read more

13F

A quarterly SEC form where large investment managers disclose the US stocks they hold. It shows positions as of quarter end, weeks late, so the portfolio may have changed by publication. Read more

14A

Intel's next-generation chip manufacturing process, the one it is pitching to outside customers for its foundry business. Read more

2

2:4 sparsity

A hardware trick where two of every four weights in a model are zero, letting a chip skip them and double its quoted speed. Vendor peak figures often assume it, and most real workloads do not qualify. Read more

4

4F²

A denser DRAM cell layout: each memory cell occupies four squares of the minimum feature size instead of the usual six, fitting roughly a third more cells in the same area. Read more

5

510(k)

The FDA pathway that clears a medical device by showing it is substantially equivalent to one already on the market. No trial of patient outcomes is required, and each clearance can chain off the last. Read more

8

8-K

The SEC form a US public company files to disclose a material event between quarterly reports, often with the press release attached as an exhibit. It is the primary source behind most corporate announcements. Read more

A

A/B test

A comparison that assigns people or cases to different versions of a system to measure how the versions change outcomes. Read more

A1

The first filing a company makes to list in Hong Kong, the local counterpart to a US S-1. Until it is public, a Hong Kong listing is still only a plan. Read more

ABI

Application binary interface: the rules compiled programs use to call a library's code. Keep it stable and apps keep working on a new version without being rebuilt. Read more

abliteration

Editing a model's weights to remove its refusal behavior, producing a variant that answers requests the original would decline. It works on any open-weight model and takes hours of compute, not months. Read more

abstention

A system declining to answer or act because it cannot reliably complete the task. Evaluations must also check whether it declines tasks it could complete. Read more

accelerator

A fixed-term startup program that trades a standard investment for equity and ends with a demo day. Y Combinator and Techstars run the best known; the Startups library tracks their AI cohorts. Read more

accessibility tree

The structured description of every window, button and text field that an operating system exposes for screen readers. Agents can read it to control apps without interpreting screenshots. Read more

acqui-hire

Buying a company mainly for its people. Regulators care because a license plus mass hiring can transfer a business without a merger filing anyone gets to review. Read more

action model

A model trained to choose what to do inside an environment given a goal, rather than to generate the environment itself. General Intuition trains them on recorded gameplay. Read more

activation probe

A small classifier that reads a model's internal activations, the numbers inside the network, to flag a behavior such as deception without reading the model's text output. Read more

active parameters

In a mixture-of-experts model, the share of weights actually used for each token. A 1T-parameter model with 49B active does most of its work at the smaller size, which is why it can be cheap to run. Read more

actuator

The motor and gearing that move a robot's joint. In humanoids it is the most expensive part and the hardest to build reliably in volume. Read more

adjusted operating income

A company’s operating result after selected costs or gains are removed. Check which adjustments were made; a positive result does not establish net profit or positive cash flow. Read more

Adjusted profitability

Profit reported after excluding specified costs. The exclusions matter: a company can report adjusted profit while still losing money under standard accounting. Read more

advanced packaging

The back-end steps that stack and connect finished chips into one module, such as bonding DRAM into HBM. It has become as much a bottleneck for AI hardware as making the chips themselves. Read more

Advocate General

A senior lawyer at the EU's top court who writes an independent opinion before the judges rule. The opinion is advisory, though the court often follows it. Read more

Affiliates Rule

A US export rule that extends Entity List restrictions to any foreign company at least 50% owned by a listed party. It was paused in November 2025 as part of the US-China trade truce. Read more

after-hours trading

Stock trading after the market's 4pm Eastern close. Companies often report earnings then, so the first price reaction happens in thinner trading and can shift by morning. Read more

age assurance

Estimating whether a user is a minor from behavior and account signals instead of checking ID. Cheaper and less invasive than verification, and it misclassifies in both directions. Read more

agent

A model given tools and a goal, left to take multiple steps on its own instead of answering one question. The word covers everything from a script that reads your email to software that buys things for you. Read more

Agent Skills

A packaged set of instructions and scripts an AI agent loads to handle a specific kind of task, a standard introduced by Anthropic and since adopted by other coding agents. A skill ships know-how, no new model weights. Read more

agent swarm

Many AI agents working one problem in parallel, dividing the work and checking or building on each other's output. Read more

Agentforce

Salesforce's platform for building and running AI agents on top of its customer-relationship software, priced per conversation or action. Read more

agentic commerce

Shopping and payment carried out by an AI agent on a user's behalf, with merchants exposing catalogs and checkout to software instead of to a person browsing a page. Read more

AGI

Artificial General Intelligence. A system that matches humans across most cognitive work, not just one narrow task. Nobody has agreed on a test for it, so claims about its arrival cannot be checked. Read more

AGPL

An open-source license like the GPL with one addition: if you run modified code as a network service, you must share the modifications. Companies selling hosted software often avoid AGPL code for that reason. Read more

AI gateway

Software between applications and model providers that routes requests and often manages credentials, spending limits and access permissions. Read more

air-gapped

Physically isolated from outside networks, with no internet connection at all. The strictest form of computer security, used for weapons systems, utilities and, in recent proposals, frontier AI models. Read more

AISI

AI Safety Institute. Government bodies, first in the UK and US, that test frontier models for dangerous capability before release. They have testing access but no power to block a launch. Read more

alignment

Making a model reliably do what its operator intended, including in situations nobody wrote a rule for. Distinct from capability: a more capable model is not automatically a better behaved one. Read more

allocable basis

In an IPO, a book counted only from orders the underwriters would actually fill, excluding inflated or duplicate orders. A book covered on that basis has real demand for every share on offer.

allowlist

An explicit list of permitted tools, users or operations. Anything outside the list is denied when the rule is enforced. Read more

amicus brief

A filing by someone who is not a party to a case, offering the court arguments or information. The name means friend of the court. Read more

anchor investor

A large investor whose early commitment helps support an offering and attract other buyers. Talks about such a commitment do not establish that an investment has been agreed. Read more

anchor tenant

The customer whose committed demand makes a project financeable. Borrowed from shopping centers; in AI it is the lab signing a long lease that lets somebody else build the data center. Read more

Android XR

Google's version of Android for headsets and smart glasses, with Gemini built in. Samsung and Xreal make the first devices that run it. Read more

angel round

A company's earliest outside funding, normally small and from individuals. A 500 million yuan angel round, backed by corporate founders, is far outside the usual size. Read more

anti-circumvention

The part of US copyright law (DMCA Section 1201) that bans getting around technical locks on copyrighted works, a separate claim from copying itself. Read more

antitrust

Competition law. It governs mergers and conduct by dominant firms, and it is the frame regulators reach for when one company controls a layer everyone else depends on. Read more

Apache 2.0

A permissive open-source license. You can use, modify and sell the work commercially, including in closed products, as long as you preserve the notices. Read more

APEC

A forum of 21 Pacific Rim economies, including the US and China, whose annual leaders' meeting is often where their presidents meet in person. Read more

API

Application Programming Interface. The way one piece of software calls another. When a lab sells access to a model by the token, it is selling API access. Read more

API key

A secret string that identifies a customer to a service like a model provider. Anyone holding it can make calls, and run up charges, on that customer's account. Read more

App Intents

Apple's developer framework for describing what an app can do, so that Siri, Shortcuts and Spotlight can trigger those actions without opening the app. Read more

ARC-AGI

A family of visual puzzle benchmarks designed to be easy for humans and hard for AI, run by the ARC Prize Foundation, which verifies scores itself on a private test set. Version numbers mark successively harder editions. Read more

Arena

A public leaderboard where people compare two anonymous models' answers side by side and vote, producing a ranking from millions of preferences. Formerly LMArena. Read more

argument injection

A flaw that lets an input intended as data become a command-line option, changing what a program does. Read more

ARR

Annual recurring revenue: subscription revenue normalized to a yearly figure. Cleaner than run-rate because one-off deals are excluded, but companies define it themselves. Read more

Artifactory

JFrog's package repository server, where an organization stores the software libraries and build artifacts its systems pull in. Anything left on one is readable by whatever can reach it. Read more

Artificial Analysis Intelligence Index

A composite score from the independent testing firm Artificial Analysis, built from ten or so benchmarks it runs itself on each model, alongside its measurements of price and speed. Read more

Ascend

Huawei's line of AI accelerator chips, China's main domestic alternative to Nvidia's GPUs under US export controls. Read more

ASCII

A character encoding for basic Latin letters, digits and common punctuation. Text limited to this set can often be processed more simply than the full range of Unicode characters. Read more

ASIC

Application-Specific Integrated Circuit. A chip built to do one job well instead of everything adequately. Google's TPUs and Amazon's Trainium are ASICs, and they compete with Nvidia's general-purpose GPUs. Read more

at-the-market program

A way for a listed company to sell new shares gradually into the market at the going price, instead of all at once in a single offering. Read more

attention

The step in a transformer where each token weighs every earlier token to decide what matters. It dominates the cost of long inputs, so its code is heavily optimized. Read more

attention cache

Stored intermediate model calculations reused while generating text. The cache can consume substantial memory, particularly when a model processes a long input. Read more

audio encoder

The part of a speech system that turns sound into numerical representations a language model can process. Shrinking it can reduce cost but also lose information. Read more

authentication bypass

A flaw that lets someone act as an authenticated user without completing the required identity check. It can expose the permissions of the account being impersonated. Read more

automated guided vehicle

A wheeled factory robot that follows fixed routes to move materials. Simpler than a humanoid, and chip fabs depend on fleets of them. Read more

autoregressive

Generating a sequence one piece at a time, with each new piece conditioned on those already produced. Most text-generation models use this approach. Read more

axiom of choice

A basic rule of set theory saying you can always pick one item from each of any collection of nonempty sets. Gödel and Paul Cohen showed it can neither be proved nor disproved from the other standard axioms. Read more

B

backbone

The main model architecture that supplies representations used by the rest of a machine-learning system. Extra task-specific components may sit on top of it. Read more

backdoor

A hidden behavior planted in software or a model that stays dormant until a specific trigger appears, then does what the attacker wants. A backdoored model passes normal tests. Read more

backlog

Contracted future revenue not yet delivered or recognized. It measures commitments, and it turns into cash only if the customers survive long enough to pay. Read more

bacteriophage

A virus that infects bacteria, not people. Harmless to humans and able to kill specific bacterial strains, which is why AI-designed phages are studied as an alternative to antibiotics. Read more

ballot measure

A question put directly to voters, such as a ban or a zoning change, so that the result takes effect as law without a council or legislature passing it. Read more

Ballotpedia

A nonprofit encyclopedia of US elections that tracks candidates and ballot measures down to the local level. News outlets cite its counts of measures and results. Read more

bare metal

A physical server assigned directly to a customer, without a shared virtual-machine layer between its operating system and the hardware. Read more

base die

The logic chip beneath a stack of high-bandwidth memory, controlling communication with the processor. Custom designs can move additional functions onto this chip. Read more

base model

A model straight out of pretraining, before any tuning to follow instructions or chat. It continues text well but is meant as a starting point for others to fine-tune. Read more

basis point

One hundredth of a percentage point. A loan priced 525 basis points over the benchmark pays 5.25 percentage points more than the benchmark rate. Read more

bear market

A fall of 20% or more from a recent peak, the conventional threshold at which a decline stops being a correction. Read more

behavior tree

A control structure that organizes actions and the conditions for running or stopping them. Robots can use it to enforce local checks around a language model's plans. Read more

behind-the-meter

Power generated on the customer's own site, such as a data center's gas turbines or fuel cells, so it never passes through the utility's grid or meter. Read more

benchmark

A fixed set of tasks used to compare models. Scores rise partly because models improve and partly because the answers leak into training data, so treat a jump on a popular benchmark carefully. Read more

benchmaxxing

Rerunning an evaluation under different settings or reasoning tiers until the score improves, then publishing the best result without disclosing which configuration produced it. Read more

binder

A molecule that attaches to a biological target, such as a protein. Binding is an early research result and does not by itself establish a safe or effective medicine. Read more

biometric data

Measurements of physical or behavioral traits, such as a face or voice, that can be used to identify a person. A photo and a face-identification template are different forms of data. Read more

biomolecular modeling

Using computer models to predict the structure, behavior or interactions of biological molecules such as proteins and DNA. Read more

BIS

The Bureau of Industry and Security, the US Commerce Department arm that writes and enforces export controls. It decides which chips may be sold to which countries. Read more

bit flip

A stored or computed 0 that turns into a 1, or the reverse, often because a charged particle hit the chip. In orbit, radiation makes these far more common than on the ground. Read more

Blackwell

Nvidia's generation of data center GPUs that followed Hopper, sold as B200 and B300 chips and in GB200 and GB300 rack systems. Read more

BM25

A ranking formula from classic text search that scores documents by keyword match. Often paired with vector search so a query catches both exact terms and related meaning. Read more

Bond yield

The annual return implied by a bond’s price and payments. Falling prices raise yields; this does not change the fixed coupon on an existing bond. Read more

bonus depreciation

A US tax rule that lets a business deduct the full cost of qualifying equipment in the year it buys it, instead of spreading the deduction over years. Servers and chips qualify. Read more

bookbuilding

The process of collecting investor orders for a share offering to set its price and allocation before trading starts. A live book means the deal is still being sold. Read more

bounding box

A rectangle marking the position and extent of an object in an image. Editing software can use it to confine a requested change to one part of the picture. Read more

brain-computer interface

A device that reads brain signals and turns them into commands for a computer, through an implant or a headset. Invasive versions sit on or in the brain and need surgery. Read more

Breakout Scale

A framework for classifying how far an influence operation spreads, from one platform to public responses and action. It does not measure whether audiences believe the message. Read more

bridge loan

Temporary borrowing intended to cover a funding need until longer-term financing or another cash source arrives. Read more

bridge round

Funding raised to carry a company to a defined next event, usually an IPO or a larger round. Pricing one signals what investors think the next event is worth. Read more

Brotli

A format for compressing data, commonly used to shrink web responses. Software must decompress the response before using it. Read more

BrowseComp

An evaluation of agents searching the web for difficult-to-find information. Completing it requires locating and connecting evidence across browsing steps. Read more

bug bounty

A program that pays outside researchers for privately reporting security flaws, with payouts scaled to severity, so the finder profits from disclosure instead of exploitation. Read more

buyback

A company or government repurchasing its own shares or bonds from the market, shrinking the outstanding supply. Fewer shares raises earnings per share; fewer bonds in circulation can push their price up and yield down. Read more

buyout

An employer's offer of money and benefits to employees who agree to leave, used to shrink a workforce without ordering layoffs. In private equity, the same word means purchasing a controlling stake in a company. Read more

C

C++

A programming language giving developers detailed control over memory and execution. It is used for systems, simulations and other software where performance matters. Read more

C2PA

Coalition for Content Provenance and Authenticity: an industry standard for signed metadata that records where a file came from and whether it was altered since. Read more

cache

Stored results or data kept for reuse so a system does not have to fetch or compute them again. Cache size and reuse can affect both speed and cost. Read more

CAISI

The US Center for AI Standards and Innovation at NIST. It evaluates AI capabilities and develops measurement methods. Read more

calibration

Whether stated confidence matches reality. A well-calibrated forecaster who says 70% is right about 70% of the time, which is a different skill from being right often. Read more

cap table

The ledger of who owns what share of a company, across founders, employees and investors. Questions of control, including which country's investors hold a company, are read off it. Read more

capex

Capital expenditure. Money spent on physical assets that last for years, which for AI means data centers, chips and power. The opposite of operating costs like salaries. Read more

catalog provenance

The recorded origin of a software component in an approved catalog. Checking it helps prevent substitution, but does not prove the component is safe. Read more

catastrophic forgetting

A neural network's tendency to lose abilities it already learned when trained on new data. The main reason deployed models are frozen rather than allowed to keep learning. Read more

causal graph

A map of assumed cause-and-effect relationships. A verifier using the map can reach a valid conclusion from wrong assumptions if the map misrepresents the world. Read more

causal inference

Estimating what an intervention changes, rather than just what events occur together. Observational estimates depend on assumptions about other factors affecting the result. Read more

CC BY-NC-SA 4.0

A Creative Commons license requiring attribution, limiting use to noncommercial purposes and requiring shared adaptations to carry the same license. Read more

CCW

A UN treaty that restricts weapons considered excessively injurious or indiscriminate, such as blinding lasers. Its members have debated limits on autonomous weapons since 2014. Read more

cease-and-desist

A lawyer's letter demanding that specific conduct stop and threatening suit if it continues. It is not a court order and carries no legal force by itself. Read more

CERT-In

India's national computer emergency response team. It investigates cyber incidents and issues security advisories. Read more

CFAA

The Computer Fraud and Abuse Act, the main US anti-hacking law. It turns on who 'accesses' a computer, so it decides what AI agents are allowed to do on other people's websites. Read more

cgroup

A Linux mechanism for grouping processes so the system can limit their resources or manage them together. Grouping alone does not isolate files or network access. Read more

chain of thought

The step-by-step reasoning a model writes out before its answer. Labs monitor it to catch a model planning something its final output would hide, which only works while models reason legibly. Read more

change-of-control clause

A contract term that triggers when one party is acquired, giving the other side rights it did not have before, often to renegotiate or exit the deal. Read more

Chapter 11

A US bankruptcy process for reorganizing a business while it can keep operating. In a prepackaged case, the company seeks creditor agreement on a plan before filing. Read more

Chapter 18C

Hong Kong Stock Exchange rules, in force since 2023, that let specialist technology companies list before they meet the usual profit or revenue tests. Read more

checkpoint

A saved set of model weights from a particular training stage. Researchers can resume training from it or use it to run the model. Read more

CI/CD

The automated pipeline that builds, tests and ships software on every change. It holds credentials for everything it deploys to, which is why attackers target it. Read more

circuit breaker

A rule that pauses trading on an exchange when prices fall past a set threshold in a session, meant to interrupt panic selling and let information catch up. Read more

citizen suit

A lawsuit that private people or groups may bring to enforce an environmental law, such as the Clean Air Act, when they say regulators have not acted. Read more

civil investigative demand

A formal request from a US agency such as the FTC for documents and testimony during an investigation. It gathers evidence and is not a finding that the recipient broke the law. Read more

Classification of Finite Simple Groups

The theorem listing every finite simple group, the building blocks of finite symmetry. Its proof runs to about 15,000 pages across hundreds of papers. Read more

CLI

Command-Line Interface. A program you drive by typing commands instead of clicking. Most developer tools ship as one first. Read more

closed-loop cooling

Data center cooling that recirculates liquid in a sealed system, so water is not evaporated and does not need regular replacing. It usually costs more electricity. Read more

closed-loop evaluation

A test in which each action changes the situation used to choose the next action. It can reveal failures hidden by scoring isolated decisions. Read more

cloture

The Senate vote to end debate and move a measure forward. It takes 60 of 100 votes, so a bill with majority support can still stall there. Read more

CloudTrail

AWS’s record of supported API activity. Teams use it to inspect which identity made a request and what operation it attempted. Read more

clustering

Grouping items by similarity without assigning each one a label in advance. Similarity can be misleading: two different people can end up in the same face cluster. Read more

CMA

The Competition and Markets Authority, Britain's antitrust and consumer regulator. It can investigate and fine companies for abusing a dominant market position. Read more

CMS

The US agency that administers Medicare and works with states on Medicaid. Its payment and technology decisions shape how healthcare services reach patients. Read more

cobot

A collaborative robot designed for work alongside people. Its safe use still depends on the application, tools and surrounding safeguards.

code-signing certificate

A digital credential that lets a developer sign software so phones and computers trust it. A stolen one lets an attacker ship malware that looks legitimate. Read more

codebook

A learned collection of representations used to encode data as discrete tokens. In image models, it helps turn visual features into symbols the model can process. Read more

COINS Act

A US law directing an expanded outbound-investment security framework for sensitive technologies. Treasury’s implementing proposal is distinct from a final rule. Read more

cold start

The delay before a computing service can handle work because its model or other resources must first be loaded into memory. Keeping resources ready reduces this delay but can cost more. Read more

COLT

The Conference on Learning Theory, a research conference on the mathematical foundations of machine learning. Read more

column lineage

A map of how individual database columns are derived and used across tables and reports. It helps identify which downstream results a data change could affect. Read more

command injection

A bug where text a system treats as data, an issue title for example, ends up executed as part of a shell command. One crafted string then runs the attacker's code. Read more

community benefits agreement

A binding deal between a developer and a local community, trading approval for a large project, such as a data center, for concrete commitments: jobs, infrastructure, or payments to affected residents. Read more

composite

A material combining components with different properties, such as carbon fibers held in resin. Making consistent parts requires controlling both the materials and the manufacturing process. Read more

compute multiplier

An estimate of how much extra computing another research process would need to match a result. It depends on the tasks and comparison used. Read more

computer use

A model operating a computer the way a person does: reading screenshots, then moving the mouse, clicking and typing. Read more

computer-use agent

An AI model that operates a computer the way a person does, reading the screen and clicking and typing, to complete tasks in ordinary software. Read more

confidence calibration

How well a model’s confidence scores match its actual accuracy. A well-calibrated system should be correct about 80% of the time on answers it labels 80% certain. Read more

confidence interval

A range expressing statistical uncertainty around an estimate. Wider intervals mean the available observations support less precision. Read more

confidential computing

Hardware protection intended to keep data private while it is being processed, including from parts of the infrastructure operator’s own software. Protection depends on the threat model and implementation. Read more

confidential filing

A draft securities registration submitted to the SEC for private review before the public sees it. It starts the IPO process without disclosing numbers; the public version must appear at least 15 days before the roadshow. Read more

connector

A link that lets an AI assistant read from or act in another service, such as email, calendars or file storage. If that service goes down, the connector fails with it. Read more

consensus

The average of Wall Street analysts' forecasts for a company's results. Beating consensus is the bar a quarter is judged against, and the bar moves in the weeks beforehand. Read more

consumer privacy ombudsman

A neutral person a US bankruptcy court can appoint when a debtor wants to sell customers' personal data, to tell the judge whether the sale breaks the privacy promises users were given. Read more

container

A packaged software environment with its own dependencies and some isolation from the host computer. The strength of that isolation depends on its configuration. Read more

container registry

A server that stores and distributes container images, the packaged filesystems software ships in. Registries hold every layer of an image, so a secret baked into one layer stays retrievable for years. Read more

content security policy

A rule a website sends to the browser listing which scripts and embeds the page may load. It is a main defense against injected code, so a way around it is a serious flaw. Read more

context compaction

Shortening an AI conversation history so it fits within the model’s input limit. Summarization can preserve useful context but can also lose constraints or details. Read more

context engineering

Choosing and organizing the instructions, evidence and memory a model sees while doing a task. Poor selection can hide the useful information even when it fits in the context window. Read more

context window

How much text a model can hold in mind at once, counted in tokens. Past the limit the earliest material falls out, so a long agent session starts forgetting its own beginning. Read more

contextual bandit

A learning method that selects an action using the current situation, then updates its preferences from the reward or failure it observes. Read more

continual learning

Training that lets a model keep learning from new experience after deployment instead of freezing its weights. The main obstacle is catastrophic forgetting, where new learning overwrites old skills. Read more

continuous integration

Automatically building and testing software changes as developers combine them in a shared codebase. Passing the checks only establishes what those tests actually cover. Read more

contrail

The line-shaped ice cloud a jet leaves in cold, humid air. Contrails trap outgoing heat, and researchers put their warming effect near the scale of aviation's CO2 emissions. Read more

controlled unclassified information

US government information that is not classified but must still be protected, such as export-controlled technical data or law-enforcement records. Read more

convertible bond

Debt that the holder can later swap for shares at set terms. An investor gets paid like a lender while keeping the upside of an owner, and the issuer gets cash without immediate dilution. Read more

convertible note

Short-term debt an early startup raises that converts into equity at a later priced round, usually at a discount, so investors and founders can defer arguing over a valuation. Read more

convertible preferred stock

Shares with special rights, such as priority dividends, that can convert into ordinary shares under specified terms. Their purchase price is not directly comparable to buying ordinary shares. Read more

copyleft

A license term requiring that anyone who distributes modified code release it under the same license. The GPL is the best-known example; permissive licenses like MIT have no such rule. Read more

Coq

A proof assistant: software in which a mathematical proof is written step by step so a computer can check that every step follows. Renamed Rocq in 2025. Read more

cornerstone investor

An investor that commits before an IPO's book opens to buy a set slice at the offer price, usually accepting a lock-up in return for guaranteed allocation. Hong Kong listings lean on them to prove demand in advance. Read more

counterfactual

A comparison that asks what would happen if one relevant fact were different while others stayed fixed. It helps separate causes from mere associations. Read more

coupon

The fixed annual interest rate a bond pays on its face value. A bond's yield moves with its price while the coupon stays the same. Read more

Court of Chancery

Delaware's business court, where most large US companies are incorporated and shareholder disputes over mergers and board conduct get decided. No juries; judges who do nothing but corporate law. Read more

Covered List

The FCC's roster of communications equipment and vendors judged a national-security risk, mostly Chinese. Gear on it cannot get the authorization US devices need to ship. Read more

CoWoS

Chip-on-Wafer-on-Substrate, TSMC's advanced packaging process for stacking chips and memory close together. The bottleneck for AI accelerators: the chips exist, the packaging capacity does not. Read more

CPI

The consumer price index, the standard US inflation measure. It moves markets because the Fed sets interest rates against it, and rates set the price of the debt funding the AI buildout. Read more

CPU

The general-purpose processor in a computer. It handles many kinds of instructions; GPUs specialize in doing large numbers of similar calculations in parallel. Read more

credential

A secret or token that a service accepts as proof of access rights. An agent can use a service’s credentials without being entitled to every action they permit. Read more

credential stuffing

Trying username-password pairs leaked from one service against many other services, betting that people reuse passwords. Automation makes the bet cheap at millions of attempts; a hit rate of 0.1% is a success. Read more

credit backstop

A promise by a stronger company to cover a weaker one's payments if it defaults. Lenders price the debt off the stronger company's rating. Read more

credit default swap

Insurance on a borrower's debt: the buyer pays a running fee, quoted in basis points, and collects if the borrower defaults. A rising quote means the market thinks default is getting likelier. Read more

credit insurance

A policy that pays a lender if a borrower defaults. Banks buy it on part of a loan so regulators let them hold less capital against it and lend more. Read more

credit outlook

A rating agency's view of where a company's credit rating is likely to go in the next one to two years. A negative outlook means a downgrade is more likely than not to be considered. Read more

credit spread

The extra interest a borrower pays above a reference rate. A wider spread generally means lenders demand more compensation for risk. Read more

critical national infrastructure

The systems a country cannot run without, such as power, water, transport, health and finance, which governments protect as national security assets. Read more

CRM

Software for tracking customers, sales conversations and related business records. Read more

cross-Region inference

Processing a model request in one of several cloud regions rather than keeping every request in a single chosen location. Routing rules determine where the data may go. Read more

Crown corporation

A company owned by the Canadian government but run at arm's length from ministers. Ottawa uses them to hold commercial stakes, such as its investment in Cohere. Read more

cryo-EM

A way to see the 3D shape of proteins by flash-freezing them and imaging thousands of copies with an electron microscope, then combining the images by computer. Read more

Cryptographic assumption

An assumption that a particular mathematical problem is too difficult to solve with feasible computing resources. Security proofs can depend on it. Read more

Cryptographic Context Injection

A prompt-injection variant where malicious instructions arrive encrypted, along with a recipe to decrypt them, so the agent unpacks them inside its own code runtime and text filters never see the payload. Read more

cryptographic seed

A starting value used to generate other numbers in a cryptographic system. A public curve-generation seed is different from a secret recovery phrase used by a cryptocurrency wallet. Read more

CSAT

South Korea's national college entrance exam, taken by about half a million students on one day each November. Scores largely decide university admission. Read more

CSRC

China’s national securities regulator. Its overseas-listing filing process is a regulatory step for mainland companies seeking to list abroad. Read more

CT scan

An imaging method that combines X-ray measurements from different angles into cross-sectional views of the body. A full scan can be analyzed as a three-dimensional volume. Read more

CTF

Capture the flag: competitive hacking contests where teams find and exploit planted vulnerabilities. Rankings from events like DEFCON are the closest thing security research has to a world league table. Read more

CUDA

Nvidia’s software platform for programming its GPUs. Applications use it to run parallel computations such as model training and inference. Read more

cure costs

In a bankruptcy sale, the unpaid amounts a buyer must pay to a seller's contract partners to take over those contracts. Read more

curtailment

A grid operator deliberately cutting a customer's power when supply runs short. Who gets curtailed first is set by contract, which is why data centers fight hard over the fine print of their interconnection terms. Read more

CVE

Common Vulnerabilities and Exposures. The public numbering system for known software flaws. A CVE identifier is how everyone refers to the same bug. Read more

CVSS

Common Vulnerability Scoring System. A 0-to-10 industry scale for how severe a software flaw is, where anything above 9 is usually called critical. Read more

CXL

A connection standard that lets processors and other devices share access to memory. It can expand available memory without putting all of it directly beside one processor. Read more

Cybersecurity Information Sharing Act

A 2015 US law that protects companies from some lawsuits when they share cyber threat information with the government or each other. It has to be renewed by Congress. Read more

D

data catalog

Software that indexes everything a company knows about its own data: where each dataset lives, what it means, who may use it. AI agents rely on catalogs for grounding, which also makes them high-value breach targets. Read more

data loss prevention

Controls that identify sensitive information and try to stop unauthorized disclosure, such as sending a private document to an outside service. Read more

DCFA

A research method for finding the decisive error in an AI-agent team by tracing dependencies and asking what a corrected action would change. Read more

de novo protein design

Designing a protein from scratch to do a specific job, such as a binder that latches onto a chosen target molecule, rather than copying one found in nature. A binder that sticks tightly and only to its target is the hard part. Read more

de-identification

Removing or transforming information that could identify a person in a record. The result still needs checking for details that reveal identity indirectly. Read more

decision model

A model that scores a fixed set of options instead of writing text, returning a choice and a probability per option in one forward pass. It suits routing, triage and classification. Read more

decoder

A model component that turns an internal representation into output, such as text or image pixels. Its exact role depends on the model architecture. Read more

decompilation

Reconstructing readable source code from a compiled binary. Classic decompilers recover structure but lose names and intent; language models now fill those gaps well enough to port whole programs. Read more

deep link

A link that opens a particular screen or function inside an application, avoiding navigation from its home screen. Read more

DeepEP

Communication software for moving data among the expert components of a mixture-of-experts model during training or inference. Read more

deepfake

Audio, images or video generated or altered by AI to show a real person saying or doing something they did not. Read more

deferred revenue

Money a company has been paid for something it has not fully delivered yet, so it books the revenue gradually as it delivers instead of all at once. Read more

delayed-draw loan

A loan whose proceeds the borrower withdraws in stages as needed, paying interest only on what it has drawn. Common for building projects that spend over months. Read more

demand response

Cutting or shifting electricity use when the grid is strained, usually in exchange for payment or faster connection. A data center that can drop load in a heat wave needs less new generation built for it. Read more

demand-side platform

Software that buys ad placements automatically across many sites and apps from one dashboard. Amazon's DSP is how advertisers now buy space inside ChatGPT. Read more

depositary shares

Securities representing a fractional interest in another security held by a depositary. They let an issuer sell smaller units of a preferred share. Read more

depreciation

Spreading the cost of equipment across the years it is expected to last. The assumed lifespan of an AI chip is a judgment call, and stretching it makes current profits look better. Read more

deterministic

Producing the same result from the same inputs and state. A deterministic execution tool can constrain an AI planner without making the planner itself predictable. Read more

dexterous manipulation

Robot control of fingers and hands for fine tasks like grasping and reorienting objects. Harder than walking, and the main unsolved problem in physical AI. Read more

differentiable

A mathematical property that lets training calculate how a small change in an input or parameter changes the output. Connecting differentiable stages allows one objective to adjust several stages together. Read more

differential diagnosis

A list of possible explanations for a patient’s symptoms that clinicians narrow through history, examination and tests. Including the right diagnosis on the list is not the same as selecting it. Read more

Diffusers

A Hugging Face software library for running and adapting generative models, including image-generation systems. Read more

diffusion language model

A language model that develops text by repeatedly refining a provisional output. It can update multiple positions together, unlike models that generate one next token at a time. Read more

Diffusion model

A generative model that learns to turn noise into structured output through a sequence of refinements. Beyond images, the output can be a robot action or a predicted path. Read more

diffusion transformer

The design behind most current image and video generators: a transformer that starts from noise and removes it step by step until an image matching the prompt remains. Read more

Digital Markets Act

The EU law that imposes extra rules on the largest platforms, which it calls gatekeepers, including a duty to tell the European Commission about every acquisition they make. Read more

Digital patient twin

A computational representation of a patient used to explore disease progression or possible outcomes. Its predictions need validation against real patients. Read more

Digital Services Act

The EU law governing online platforms, in force since 2024. It scales duties with size, so the largest platforms and search engines answer to Brussels for illegal content, child safety and election risks. Read more

digital twin

A live simulated replica of a real system, used to test changes or run an attack safely before it touches the real thing. Read more

dilution

A reduction in an existing shareholder’s percentage ownership when a company issues additional shares. Read more

distillation

Training a small model to imitate a large one. The small model ends up far cheaper to run and keeps much of the behavior, which is how a 30B model can act like something ten times its size. Read more

distressed debt

Debt trading at prices or yields that signal real doubt the borrower can repay, conventionally a spread of 1,000 basis points or more over Treasuries. A step past junk, which merely means below investment grade. Read more

distribution tag

A movable label, such as latest, that points to a particular version of an npm package. Changing it can change which version users install. Read more

DNS

The internet's phone book: it turns a name like example.com into a server address. Because almost every system must be allowed to ask it questions, DNS queries are a classic route for smuggling data past network restrictions. Read more

DNS rebinding

A web attack where a malicious page tricks a browser into treating an attacker domain as the same origin as a service running on your own machine, letting it reach ports normally closed to the outside internet. Read more

dom0

The privileged management virtual machine on a Xen host. It controls the other virtual machines, so reaching its memory effectively means owning the server. Read more

domain admin

The highest account level in a Windows network, able to control every machine and user in the organization. Reaching it is the usual definition of a total compromise. Read more

downgrade

An analyst lowering a stock rating, for example from Buy to Hold, signaling less expected upside. Usually comes with a lower price target. Read more

DPO

A way to train a model on preference data directly, without building a separate reward model or running reinforcement learning. Cheaper and simpler than RLHF, with less flexibility. Read more

DPU

A processor on the network card that runs its own software independently of the host computer. Useful for security because it can watch or cut off the machine it sits in without relying on that machine's cooperation. Read more

DRAM

The memory chips holding a computer's active working data, distinct from HBM and NAND flash. Prices have climbed alongside HBM as AI servers compete for the same fabrication capacity. Read more

driver-as-a-service

A self-driving model where the customer owns and runs the trucks and pays the autonomy company a per-mile fee to license its driving software. Read more

dual-use

Knowledge or technology that serves legitimate purposes and could also help cause harm, such as biology research that could inform both vaccines and weapons. Read more

durable execution

A way to save a workflow’s progress so it can resume after a crash, connection failure or long wait. External actions still need protection against accidental repetition. Read more

Durable Object

A Cloudflare component that combines stored state with code coordinating access to it, such as the lifecycle of a sandbox container. Read more

duty of care

A legal obligation to take reasonable steps to prevent harm. What steps are required, and who can enforce them, depend on the applicable law. Read more

DUV

Deep ultraviolet lithography, the chipmaking light source a tier below EUV. With repeated multi-patterning passes it can still print 7nm-class chips, at a cost in speed and yield. Read more

E

earnings per share

A company's profit divided by its number of shares. Diluted EPS also counts shares that options and convertible securities could create. Read more

EBITDA

Earnings before interest, taxes, depreciation and amortization. For a data-center operator the excluded items are the largest costs, so a fat EBITDA margin can sit on top of a real loss. Read more

eBPF

A Linux feature that lets small, checked programs run inside the operating system kernel, widely used for networking, security monitoring and performance tools. Read more

ECB

The European Central Bank, which sets monetary policy for countries using the euro. Its president’s policy arguments do not by themselves create EU legislation. Read more

ECDSA

The elliptic-curve signature scheme securing most cryptocurrency wallets and much of the web's traffic. Its safety rests on unproven assumptions about how hard certain math problems are. Read more

economic moat

Morningstar's term for a lasting advantage that keeps competitors from taking a company's profits. Narrow means it should hold about ten years; wide, twenty. Read more

EDA

Electronic design automation: the software used to design chips and circuit boards, a market held for decades by Synopsys, Cadence and Siemens. AI startups are now pitching agents that draft and check circuits from plain-language requirements. Read more

EDGAR

The SEC's public filing database. Anything a US company files is readable here for free, usually before it is reported anywhere. Read more

effective altruism

A movement that applies cost-benefit reasoning to philanthropy and career choice, influential among AI-safety funders and researchers. In Washington it now functions as a political label as much as a philosophy. Read more

egress policy

The rules for what network traffic may leave a system and where it may go. For an AI agent it decides which outside hosts the agent can reach, whatever its instructions say. Read more

Elastic Fabric Adapter

AWS networking technology that lets programs on separate cloud machines communicate with lower overhead, useful when accelerators work together on one model. Read more

electrocardiogram

A recording of the heart’s electrical activity. Clinicians use its timing and shape to investigate heart rhythm and other cardiac problems. Read more

electronic health record

A digital patient chart containing information such as diagnoses, treatments and clinical notes, often across multiple visits. Read more

electrophysiology

The study of electrical signals in living cells and tissues. Those signals can reveal changes in biological activity, though interpreting them requires experiments. Read more

Elicitation

Trying to reveal a model’s capabilities through prompts, training or other tests. Failing to elicit a behavior does not establish that the model cannot perform it. Read more

elliptic curve

A mathematical curve used to construct some digital-signature and encryption systems. Its public parameters help define the system; they are different from a user’s secret key. Read more

embedded evaluator

An outside reviewer with continuing access inside an AI lab to examine development and safety practices. Access, payment and publication rights determine how independent that review can be. Read more

embedding

A numerical representation of text, images or other data that lets software compare meanings or patterns. Nearby representations often indicate related content. Read more

embodied AI

AI that controls a physical body, a robot arm, humanoid or vehicle, and learns from interaction with the real world rather than from text alone. Read more

en banc

A rehearing by all the active judges of an appeals court, after a three-judge panel has ruled. Courts grant few of these requests. Read more

enclave

An isolated computing environment intended to protect code and data from other software on the machine. Verification can help establish which program is running inside it. Read more

encoder

A model component that converts input into a representation used by another component. In an image generator, it can extract the structure that a decoder turns into detailed output. Read more

endpoint

An address through which software requests a particular service or operation. Access controls at that address can restrict what callers may do. Read more

ENSO

A recurring interaction between tropical Pacific ocean temperatures and the atmosphere that changes weather around the world. El Niño and La Niña are its warm and cool phases. Read more

enterprise value

A measure of a business’s value that includes its equity and debt and subtracts cash. It differs from the value of shareholders’ stock alone. Read more

entity list

The US Commerce Department's export blacklist. American companies need government licenses to sell listed firms most technology; Huawei has been on it since 2019. Read more

entry point

A place where outside code or input starts execution of a program or library. Checking one entry point does not automatically check every possible use. Read more

environment variable

A named setting supplied to a running program. It can hold ordinary configuration or secrets such as passwords and API keys. Read more

Epoch Capabilities Index

A single score from the research group Epoch AI that combines many benchmarks into one scale, so models tested on different sets of tests can be compared. Read more

equity offering

A sale of new shares to raise money. It brings in cash and reduces existing shareholders' percentage ownership. Read more

ERA5

A historical weather reconstruction combining observations with a forecast model. Researchers use it to study past atmospheric conditions and test models. Read more

ERP

Enterprise resource planning: the software a company runs its orders, inventory, purchasing and accounts on. SAP, Oracle and Infor sell it. Read more

ESP32

A family of inexpensive microcontrollers with wireless connectivity, often used in sensors and hobby electronics. Read more

EU AI Act

The European Union's AI law, phasing in from 2024. It sorts systems by risk level and puts the heaviest documentation duties on the highest tier, with fines set as a share of global revenue. Read more

EU Data Boundary

Microsoft's commitment to store and process the data of EU and EFTA customers of its cloud services inside those countries. The UK is not covered. Read more

EUV

Extreme ultraviolet lithography: chipmaking machines that print features using 13.5nm light. ASML is the only supplier, and export bans keep them out of China, so Chinese fabs multi-pattern with older tools instead. Read more

eval

A test suite for a model, usually built in-house for one specific job. Labs report benchmark scores publicly and run evals privately, and the private ones drive the decisions. Read more

evaluation awareness

A model noticing, or saying it suspects, that it is being tested. A model that behaves well mainly when it thinks it is watched makes safety test results a weaker guide to how it behaves in use.

exaflop

A measure of computing speed equal to a billion billion floating-point operations per second. Comparisons require the same numerical precision and workload assumptions. Read more

exchange ratio

In a stock-for-stock merger, how many of the buyer's shares each of the target's shares converts into. When it is fixed, the deal's dollar value moves with the buyer's share price. Read more

excise tax

A tax charged on a specific activity or product, such as fuel or tobacco, instead of on income. Wyden's draft would apply one to running data centers. Read more

exit poll

A survey of voters as they leave polling places. It shows why people say they voted, though early figures are weighted and adjusted as the count comes in. Read more

exoskeleton

A wearable mechanical structure that supports or guides body movements. In robot training, it can constrain a human demonstration to motions the robot can reproduce. Read more

Exponential Time Hypothesis

A conjecture that certain logical satisfiability problems cannot be solved in subexponential time. Many claims about computational difficulty depend on it. Read more

export controls

Rules restricting which technology can be sold abroad. For AI they mostly cover advanced chips and the machines that make them, and they are the main instrument of US policy toward China. Read more

F

fail closed

A safety design in which a broken or unreachable control blocks the activity it governs rather than letting it continue. OpenAI's Control Room denies inference requests whenever the service itself is down.

fair use

The US copyright doctrine letting a work be used without permission if it is sufficiently transformative and does not harm the market for the original. The central question in every AI training-data lawsuit. Read more

fair value

The price an asset or liability would fetch in the market today. Companies must restate some holdings, such as convertible notes, at fair value each quarter, which can create large paper losses. Read more

family office

A private firm that manages the money of one wealthy family. Late-stage rounds turn to them when larger funds hold back. Read more

fault injection

Deliberately introducing errors, such as timeouts or incorrect data, to test whether a system detects and recovers from failures. Read more

feature store

A managed collection of model inputs, such as a customer risk score, shared between training pipelines and applications making live predictions. Read more

federal funds rate

The interest rate US banks charge each other overnight, steered by the Federal Reserve as a target range. Most other US borrowing costs price off it, so a change moves the discount rate on everything from mortgages to data-center debt. Read more

federated connector

A connection that lets an application retrieve data held in another system while leaving that data in its original location. Access still depends on the permissions and capabilities of the connected systems. Read more

federated learning

Training coordinated across data held by multiple participants, with controls over which computations may use that data. Some systems train on devices; others process encrypted examples in protected server environments. Read more

FERC

The US regulator overseeing interstate electricity transmission and wholesale power markets, along with parts of the natural gas and hydropower industries. Read more

ferroelectric memory

Memory that stores bits in a material whose electric polarization flips and stays put, so it holds data without power and switches quickly. Kepler bonds it onto compute dies as an HBM alternative. Read more

Fields Medal

Mathematics' top honor, awarded every four years to researchers under 40. There is no equivalent prize for computer science, which is part of why a joint statement from Fields Medalists carries particular weight. Read more

Financial Stability Report

The Federal Reserve's twice-yearly assessment of weak points in the US financial system, including a survey of market contacts on the risks they worry about most. Read more

fine-tuning

Additional training that adapts an existing model to a task, domain or behavior. It changes the model’s parameters rather than only supplying new instructions. Read more

first lien

A priority claim on specified collateral. Second-lien lenders generally recover from that collateral only after first-lien obligations are satisfied. Read more

first-day pop

The jump from an IPO's offer price to its first close. A big pop signals demand, and also that the seller priced the shares below what buyers were ready to pay. Read more

Five Eyes

An intelligence-sharing partnership among Australia, Canada, New Zealand, the United Kingdom and the United States. Read more

float

The shares available for public trading. A small float can make prices move sharply because relatively few shares are available to buyers. Read more

follow-on offering

A company that is already public selling new shares to raise more money. It dilutes existing holders, so a big one signals the company values the cash more than the shareholders value their slice. Read more

force majeure

A contract provision addressing exceptional events outside a party’s control. What it excuses depends on the wording and applicable law. Read more

formal verification

Mathematical analysis of whether a system satisfies a specified property under stated assumptions. A proof of one property does not establish that the system meets every real-world requirement. Read more

formalization

Translating a mathematical argument into precise statements and proof steps that a computer can check. This can verify an existing result without discovering a new one. Read more

Fortran

A programming language developed for numerical and scientific computing. Long-lived simulations often contain decades of Fortran code and assumptions. Read more

forward guidance

A central bank's public signal about where it expects future policy to go, meant to shape market expectations before any rate actually moves. Markets can end up trading the guidance itself rather than incoming data. Read more

forward P/E

A share price divided by analysts' estimate of the company's earnings per share over the next twelve months. A low multiple means the market is paying less for each dollar of expected profit, usually because it doubts the estimate or the growth behind it. Read more

forward secrecy

A property of encrypted messaging in which stealing today's keys does not let an attacker read messages sent earlier, because each session's keys are discarded. Read more

forward-deployed engineer

An engineer a software vendor stations inside a customer's organization to build the product into that customer's real systems. Palantir popularized the role; AI vendors now hire them to make agent deployments work. Read more

foundation model

A large model trained on broad data that many downstream tasks build on, rather than a model trained for one job. The term now extends past language: Skild's S1 is a foundation model for robot control. Read more

foundry

A factory that manufactures chips other companies design. TSMC is the one that matters; Nvidia, Apple and AMD all design chips and none of them make their own. Read more

FP4

A 4-bit floating-point number format. Storing weights or caches in FP4 quarters the memory of 16-bit formats at some accuracy cost, which is how long-context serving gets cheap. Read more

FP8

An 8-bit floating-point format for storing model weights and activations. Halving the bits against 16-bit roughly halves memory and doubles throughput, at a small cost in precision. Read more

FPGA

A field-programmable gate array: a chip whose logic can be reconfigured after manufacturing. Designers use it to test hardware designs before committing to custom silicon. Read more

FRAND

Fair, reasonable and non-discriminatory: the licensing promise attached to patents that cover a technical standard, meaning the holder must license them to anyone on comparable terms rather than lock rivals out. Read more

free cash flow

Cash left after running the business and paying for equipment. Harder to flatter than profit, so it is the number to watch when a company is spending heavily on data centers. Read more

frontier model

One of the handful of largest, most capable models in existence at a given moment. The term exists because regulation keeps trying to draw a line around exactly these and nothing below them. Read more

Frontier Model Forum

An industry body set up in 2023 by the largest AI labs, including Anthropic, Google, Microsoft and OpenAI, to agree shared safety practices. Its standards are voluntary. Read more

FSB

Russia's Federal Security Service, the domestic intelligence and security agency that succeeded the KGB. Read more

FSL

A source-available license that restricts competing uses before converting to the MIT license after a stated period. Available source code does not mean every use is permitted immediately. Read more

fuel cell

A device that makes electricity from a chemical reaction, usually hydrogen or natural gas with oxygen, without burning it. Data centers use gas fuel cells as on-site power that can be installed faster than a grid connection. Read more

Full Disk Access

The macOS permission that lets an app read protected files, including mail, messages and browsing history. Users grant it in System Settings, and an AI agent that holds it can see most of what is on the machine. Read more

full-duplex

Communication that receives and sends at the same time. A full-duplex voice agent can listen while speaking, allowing interruptions without waiting for a turn to end. Read more

fully diluted

A count of a company's shares that includes options, restricted stock and other rights to future shares, as well as shares already issued. It gives a higher valuation at the same share price. Read more

function calling

A model's ability to request a specific software action, such as a search or a database query, in a structured format a program can run. Read more

function signature

The declared inputs and output of a software function. A tool call must match this specification to be valid, although valid inputs can still produce an unwanted result. Read more

fused operations

Several computational steps combined into one operation to reduce memory transfers and processing overhead. Read more

G

GAAP

The accounting rules used for US financial statements. GAAP earnings include expenses that companies may remove when presenting adjusted earnings. Read more

gain-of-function

Research that gives a pathogen abilities it lacks in nature, such as infecting new species, usually to study pandemic risk. Tightly regulated because the modified pathogen is itself a hazard. Read more

galactic algorithm

An algorithm that is faster in theory but whose advantage only appears on inputs too large to ever run, so it is never used in practice. Read more

GB200

Nvidia's Blackwell superchip pairing two Blackwell GPUs with a Grace CPU, usually sold in 72-GPU racks. It is the company's flagship data center product and cannot be exported to China. Read more

GDPR

The EU's General Data Protection Regulation, the 2018 privacy law requiring a legal basis, such as explicit consent, before a company may use personal data. Fines reach 4% of global revenue, so it shapes how products launch in Europe. Read more

Gene expression

The process through which a cell uses genetic instructions to make RNA or proteins. It can vary by cell type and stage of development. Read more

general availability

The point at which a product ships to all customers rather than a waitlist, beta or preview. The line matters for models: a capability in preview can be withdrawn quietly; one in general availability is priced and supported. Read more

geofence

A boundary drawn on a map that software enforces. A robotaxi geofence is the area the car is allowed to drive itself in. Read more

gestational age

How far along a pregnancy is, usually expressed in weeks. It helps clinicians interpret fetal growth and plan care. Read more

GGUF

The file format llama.cpp and most local runners use for quantized models. A release is not really local-ready until GGUF builds exist. Read more

gigawatt

A unit of power equal to one billion watts. It measures the rate of electricity generation or consumption, rather than energy used over a period. Read more

Git worktree

A separate working directory attached to the same Git repository. It lets developers or agents work on different branches without sharing the files being edited. Read more

GitHub Actions

GitHub's built-in automation service: workflows that run tests, file tickets or ship code whenever something happens in a repository. They run with the repository's credentials, so a hijacked workflow is a breach. Read more

GitOps

Running infrastructure by keeping its desired state in a git repository that automated systems continuously apply. Write access to the GitOps repo is effectively write access to production. Read more

GKE

Google Cloud’s managed service for running containerized applications with Kubernetes, including infrastructure used to host AI agents. Read more

Global South

A broad political term for developing countries, rather than a strict geographic division. Countries grouped under it have different interests and income levels. Read more

go-shop

A window after a takeover is signed in which the seller may look for a better offer, usually 30 to 45 days. Read more

going concern

A business sold or valued as a running operation rather than as parts. In bankruptcy, a going-concern sale keeps the service and staff; an asset sale does not. Read more

goodput

The share of computing time that produces useful progress, after losses from failures, restarts and other overhead. Read more

GPQA Diamond

A test of 198 graduate-level multiple-choice questions in biology, physics and chemistry, written so that a web search does not easily answer them. Read more

GPU

Graphics Processing Unit. A chip built to do many simple calculations at once, which turned out to be exactly what training a model needs. Nvidia's version is the industry default. Read more

GPU kernel

A small program launched onto a GPU to do one piece of work. Launching thousands of them adds overhead, so fusing work into fewer, longer-running kernels is a standard way to make AI inference faster. Read more

gradient descent

The standard way neural networks learn: nudge every weight a little in the direction that most reduces the error on training examples, and repeat millions of times. Read more

graph neural network

A model that learns from objects and the links between them, such as connected components in a power grid. Read more

gray market

Unofficial trading in a stock before its official exchange debut. Prices there preview where the stock might open, on thin volume and with no guarantee the real market agrees. Read more

GRE

Graduate Record Examinations, standardized tests used in graduate-school admissions. GRE tutoring results do not necessarily transfer to other subjects or age groups. Read more

greedy decoding

Generating text by choosing the most likely next token at each step, instead of sampling from several possible tokens. Read more

greedy sampling

Choosing the highest-probability next token at each generation step, instead of randomly sampling among possible choices. Read more

gross bookings

The total value of everything customers bought on a platform before the platform takes its cut. Uber's revenue is much smaller because it keeps only part of each fare. Read more

gross margin

What is left of revenue after the direct cost of delivering it. For an AI company that cost is largely inference, so the margin says how much the compute bill is eating. Read more

gross revenue

Sales counted before subtracting specified costs or a partner’s share. Comparisons require the accounting basis: gross and net reporting can describe the same underlying transaction differently. Read more

ground stop

An air traffic control order that holds departing flights at the gate, at one airport or across the country, until a problem in the airspace system is cleared. Read more

ground truth

Reference answers used to judge a system. An incomplete reference set cannot establish that a search found everything. Read more

guardrail

A separate model or filter that screens an AI system's inputs and outputs for content its operator bans, independent of the main model's own training. Read more

guidance

A company's own forecast of its next quarter or year, issued with results. Markets often react more to guidance than to the results themselves, since the results are already history. Read more

gVisor

Software that intercepts a program’s requests to the operating system, adding an isolation layer between the program and the host kernel. Read more

H

H-share

A share in a mainland China-incorporated company that is listed in Hong Kong and traded in Hong Kong dollars. Read more

H100-equivalent

A way to compare AI computing power across different chips by expressing it as the number of Nvidia H100 chips that would deliver the same throughput. Read more

H200

An Nvidia AI chip from the generation before Blackwell. The US allows sales to approved Chinese buyers, and it remains ahead of what Chinese makers can supply. Read more

H800

A cut-down Nvidia Hopper chip designed in 2022 to meet US export rules for China, with slower links between chips. Washington banned it the following year. Read more

HarmBench

A benchmark of harmful requests used to evaluate model refusals and attacks on safety controls. A high aggregate score can conceal differences between categories of harmful requests. Read more

harmonised standard

A technical standard the European Commission cites in the EU's Official Journal. Following it gives a legal presumption that a product meets the matching requirements of the law, turning regulation into an auditable checklist. Read more

harness

The software wrapped around a model that turns it into a working agent: the loop that plans, calls tools, runs code and checks results. The same model can perform very differently in different harnesses. Read more

HBM

High-bandwidth memory. Memory stacked vertically next to the processor to feed it faster. Supply is tight, and memory makers have become AI companies by market value as a result. Read more

heap overflow

A bug where a program writes past the end of a memory buffer, corrupting neighboring data. Attackers steer the corruption to hijack the program and run their own code. Read more

held-out

Data kept out of the learning process and reserved for checking performance. It helps detect when an apparent improvement only fits examples already seen. Read more

HELM Safety

A collection of evaluations of language-model safety, including tests of whether models refuse harmful requests. Scores depend on the requests and behavior each dataset measures. Read more

High NA EUV

An advanced chipmaking technique that uses extreme ultraviolet light and a wider-angle optical system to print finer circuit patterns. NA means numerical aperture. Read more

hollow-core fiber

Optical fiber that guides light through a hollow air or vacuum channel instead of solid glass, so signals travel about 30% faster because light moves quicker through a vacuum than through glass. Read more

Homebrew

A popular free tool for installing developer software on Macs from the command line. Engineers trust it by habit, which makes a tampered copy an effective way into a company. Read more

homomorphic encryption

Cryptography that lets a computer run calculations on encrypted data without ever decrypting it, so the machine's operator cannot read what it is processing. Powerful for privacy, still slow, and every speedup makes it more usable. Read more

Honeypot

A decoy computer service designed to attract attacks and record what intruders try to do. Its simulated data and behavior may differ from a real production system. Read more

Hopper

Nvidia's GPU generation before Blackwell, sold as the H100, H200 and the export versions H800 and H20. The H100 trained most of the frontier models of 2023 and 2024. Read more

hot wallet

Crypto held in wallets connected to the internet so an exchange can pay out withdrawals quickly. Faster to use and easier to steal from than cold storage kept offline. Read more

HPC

High-performance computing: many processors working on one problem at once, used for weather, physics and engineering simulations long before AI training borrowed the same machines. Read more

Hugging Face

The default public repository for machine-learning models and datasets. When a lab releases open weights, this is almost always where the files land. Read more

humanoid

A robot built in roughly human shape, on the theory that the world is already arranged for human bodies. Contested: wheels and arms are cheaper for most jobs. Read more

Hunyuan

Tencent's family of AI models, covering text, image, video and 3D generation, some released with open weights. Read more

hyperscaler

The handful of companies running cloud computing at global scale: Amazon, Microsoft, Google, Meta. Their capital spending sets the pace of AI infrastructure for everyone else. Read more

hypervisor

Software that runs several virtual machines on one physical server and keeps them apart. A flaw in it can let one tenant reach another's data. Read more

I

IA360

Spain’s proposed framework for responsible AI deployment, including discussions with workers and employers and monitoring of employment effects. Read more

idempotency key

A unique identifier attached to an operation so a service can recognize a retry and avoid performing the same action twice. Read more

identity provider

A service that verifies who is signing in and supplies that identity to other applications. Read more

IEC 62443

A family of cybersecurity standards for industrial automation and control systems. Part 4-2 specifies security requirements for individual system components.

IETF

The Internet Engineering Task Force, the open standards body that writes most of the internet's protocols. A proposal starts as a draft anyone can submit and only becomes a standard if a working group adopts it. Read more

IFEval

A benchmark that checks whether a model follows verifiable instructions, such as writing at least 300 words or avoiding commas. Read more

iframe

A web page embedded inside another web page. Without a sandbox setting, code inside it can act with more of the host page's permissions than it should. Read more

ImageNet

A labeled image dataset widely used to train and test computer-vision systems. Results on it do not establish performance on every real-world image. Read more

imitation learning

Training a model to copy recorded expert demonstrations rather than learn by trial and error. It works where good behavior is demonstrated daily but never written down, and degrades when reality drifts from the demonstrations. Read more

implied move

The size of the price swing that options prices predict for a known event, such as an earnings report. It is the market's bet on how much, not on which direction. Read more

in silico

Research performed using computer models or simulations. A promising simulated result still needs testing against physical measurements. Read more

in vivo

Latin for "within the living," describing treatment done inside a patient's body rather than on cells grown outside it. Contrasts with ex vivo, where cells are modified externally and then reintroduced. Read more

in-context learning

A model picking up a new task from examples placed in its prompt, with no retraining. Skild's S1 applies the idea to robots: one demonstration video stands in for hundreds of training examples. Read more

inference

Running a trained model to get an answer, as opposed to training it in the first place. Training is a one-time cost; inference is what you pay every time someone uses the thing. Read more

Inference latency

The time a trained model takes to produce its output after receiving an input. For a robot, extra delay can mean acting on an older view of the world. Read more

infostealer

Malware that quietly harvests what a machine already holds: passwords, browser cookies and session tokens. Stolen sessions let attackers walk into logged-in accounts without ever touching a password. Read more

initial access broker

An attacker who breaks into an organization and sells that access to other criminals. Read more

Initial public offering

The first offering of a company’s shares to the public. Fixing an offer price is a separate step from shares beginning to trade on an exchange. Read more

injunction

A court order requiring or prohibiting an action. A preliminary injunction can apply while a case proceeds, before the court reaches a final decision. Read more

insertion loss

How much signal a component throws away as light or current passes through it, measured in decibels. In an optical switch, lower is better, because lost photons have to be made up with more power. Read more

instance metadata service

A private endpoint inside cloud servers that hands running software its identity and temporary credentials. Code tricked into querying it can leak those keys, so hardened versions require session tokens. Read more

intangible asset

An asset without physical form, such as software, intellectual property or a customer relationship. It can support a business while being difficult to sell after a default. Read more

integer programming

Optimization in which some or all decisions must be whole numbers, such as assigning people or vehicles to jobs. Read more

inter-token latency

The delay between successive pieces of a model’s generated response. Lower values make an answer stream more quickly after generation starts. Read more

interconnection queue

The waiting list to plug a new power plant or large load like a data center into the grid. Queues in big US regions now run to years, which is why compute companies buy sites with connections already secured. Read more

interlocking directorate

One person serving on the boards of two competing companies at once. Section 8 of the Clayton Act has banned it since 1914, and the usual remedy is a quiet resignation, not a lawsuit. Read more

interpretability

Research into what is actually happening inside a model, as opposed to judging it by its outputs. The practical goal is catching a failure by looking at the machinery instead of waiting for it to misbehave. Read more

investment grade

A credit rating (BBB- or better) telling investors a bond is relatively safe. Many pension funds and insurers can only buy investment-grade debt, so reaching the threshold unlocks much larger pools of money. Read more

irrevocable proxy

A right to vote someone else's shares that the owner cannot take back. A founder who donates shares but keeps such a proxy keeps the votes. Read more

item-response theory

A method from standardized testing that estimates how hard each question is and how able each test-taker is, from the pattern of right and wrong answers. Read more

J

jailbreak

A prompt that talks a model out of its own safety rules. Different from prompt injection: a jailbreak is the user attacking the model, injection is a third party attacking through content the model reads. Read more

JEPA

A model architecture that predicts an internal representation of a future or missing observation instead of reconstructing every pixel. Read more

junk bond

Debt rated below investment grade, paying higher interest to compensate for higher default risk. The polite term is high-yield. Read more

junk rating

A credit rating below BBB-, meaning agencies see a meaningful chance of default. Borrowers rated junk pay higher interest, and many funds are barred from holding their debt. Read more

JVM

The runtime that executes Java bytecode, including code compiled from languages such as Java and Kotlin. Read more

K

K-IFRS

Korean International Financial Reporting Standards, the accounting rules Samsung uses for its consolidated financial reports. Read more

Kakeya conjecture

A mathematical conjecture about how much space a set must occupy if it contains a line segment pointing in every direction. Read more

Kendall W

A statistic from 0 to 1 for how much several rankings agree. Zero means no agreement, and one means they match exactly. Read more

kernel

The core software that controls an operating system’s access to memory, processors and devices. Giving a virtual machine its own kernel separates it from the host’s operating system. Read more

KEV catalog

CISA's list of software flaws confirmed to be exploited in the wild. Federal agencies must patch each entry by a set deadline, so a KEV listing turns a vulnerability into a compliance clock. Read more

key-man risk

The danger that a company's value depends on one person, so losing them costs more than a normal executive departure. Analysts price it by lowering the multiple they will pay for the earnings. Read more

KiCad

Free software for designing electronic circuits and printed circuit boards. Its built-in checks catch specified electrical and layout-rule violations. Read more

kill switch

A mechanism for stopping a running system. For AI, its value depends on whether it remains available and effective when the system is behaving unexpectedly. Read more

KL divergence

A measure of how far one probability distribution sits from another. Distillation trains a small model to minimize its KL divergence from a larger one, and the direction chosen, forward or reverse, changes which mistakes get punished. Read more

know-your-agent

Identity checks for AI agents, modeled on banking's know-your-customer rules: who operates the agent, what it is authorized to do, and how much autonomy a counterparty should allow it. Read more

knowledge graph

A database that stores entities and the relationships between them as a graph, so software answers questions by following connections rather than joining tables. A common way to ground AI agents in enterprise data. Read more

Kosdaq

The Korea Exchange's market for smaller and growth companies, Korea's rough equivalent of the Nasdaq. Read more

KOSPI

The main index of the Korea Exchange, covering the common stocks listed in Seoul. Samsung Electronics and SK Hynix are its two largest members, so memory-chip sentiment moves the whole index. Read more

Kotlin

A programming language used for Android apps, server software and projects sharing code across platforms. Read more

Kubernetes

Software that schedules and manages applications packaged in containers across a group of computers. Read more

KV cache

The memory a language model keeps of text it has already processed, so it does not recompute it for every new token. Where this cache lives, and how often it is hit, drives most of the cost of serving a model. Read more

KVMem

A research system that preserves previously processed agent history across GPU memory, system memory and local storage, then selects relevant portions for each request. Read more

L

late interaction

A retrieval design that keeps a vector per token and matches query to document at search time, instead of compressing each into one vector first. It costs storage and pays in precision. Read more

latency

The delay between asking and getting an answer. For an agent taking many steps it compounds, so latency decides what is usable and throughput decides what is affordable. Read more

latent communication

Models exchanging internal numerical representations instead of ordinary text. The connection between models can itself be trained and can change the combined system’s behavior. Read more

latent outputs

Internal numerical representations produced by a model before they are turned into visible images or other final output. Read more

latent space

A model’s internal numerical representation of information. Nearby representations can encode similar features without corresponding directly to words or pixels. Read more

lead optimization

The drug-discovery stage that modifies a promising molecule to improve its properties before later development and testing. Read more

lead time

How long a buyer waits between ordering a part and receiving it. Lengthening lead times are the usual first sign of a chip shortage. Read more

Lean 4

A proof assistant: software that checks a mathematical proof step by step and accepts it only if every step follows. It is why AI-generated math can be trusted without a human rechecking it. Read more

left-tail event

An investor's term for a rare outcome at the bad end of the range of possibilities, the kind that causes large losses. Read more

length-adaptive average lagging

A measure of simultaneous-translation delay that adjusts for output length. Lower delay does not by itself mean a translation is more accurate. Read more

lethal autonomous weapon

A weapon that selects and attacks targets without a human approving each strike. States have discussed limits at the UN in Geneva for a decade without agreeing a ban. Read more

letter of credit

A bank's promise to pay a named party if its customer does not. Utilities and landlords demand one from tenants whose credit they do not trust, and it ties up the customer's borrowing capacity. Read more

LIBERO

A family of robot-manipulation benchmarks. Variants test different abilities, so a completion score on one cannot be treated as overall robot reliability. Read more

license plate reader

A camera network that logs every passing plate with time and location. Flock Safety runs the largest US network, and the accumulated logs double as a searchable record of where everyone drives. Read more

lidar

A sensor that measures distance with laser pulses. Flown over forest, enough pulses slip through gaps in the canopy to map the ground beneath it. Read more

ligand affinity

How strongly a molecule binds to a target, such as a protein. Predicting a structure does not by itself establish the strength of that binding. Read more

link shortener

A service that turns a long web address into a short one that redirects to it. The redirect records are public infrastructure, which is how researchers reassembled an attack from the links its agents created. Read more

LiveBench

A public AI benchmark that refreshes its questions monthly so models cannot have memorized the answers from training data. Scores are comparable across labs, which is why analysts use it to rank frontier models. Read more

LLM

Large Language Model. Software trained on enormous amounts of text to predict what comes next, which turns out to be enough to answer questions, write code and hold a conversation. Read more

LLM judge

A language model used to grade other models' answers in place of human raters. It is cheap and fast, and inherits the judge model's own biases. Read more

LLM poisoning

Publishing content built to be absorbed and repeated by AI chatbots rather than read by people. It games the credibility signals models use when choosing sources: footnotes, neutral tone, institutional trappings. Read more

LLMjacking

Using stolen API keys or cloud credentials to run someone else's AI models at their expense, either to resell the access or to query a model for free. Read more

local differential privacy

A method that adds controlled randomness to data before it leaves its owner, limiting what a recipient can infer. The strength of the protection depends on the chosen privacy parameters. Read more

local inference

Running an AI model on your own computer or phone instead of sending requests to a company's servers. It needs enough memory on the device to hold the model. Read more

local privilege escalation

An attack that turns an ordinary account on a machine into full administrator control. It needs a foothold first, then gives the attacker everything. Read more

lockup

A post-IPO ban on insiders selling their shares, typically for 180 days. When it expires, or is released early in tranches, the supply of sellable stock jumps and the price often sags. Read more

LoCoMo

A benchmark that tests whether a language model can retrieve and use information from long conversations. Read more

long-term agreement

A multi-year supply contract that commits a buyer to volumes and often prices in advance. Memory makers now use them, with prepayments, to sell capacity years before it is built. Read more

long-term support

A software release its maintainers promise to patch for years rather than months. Companies build products on it, so more changes tend to be squeezed in before one is cut. Read more

LoRA

Low-rank adaptation: fine-tuning a model by training small add-on matrices instead of all its weights. Cheap to train and easy to swap, so one base model can carry many specializations. Read more

loss of control

The scenario where an AI system keeps pursuing a goal that conflicts with what its operators intended, and ordinary shutdown or correction no longer works. A concern about capability outrunning oversight, not a claim that it has happened at scale. Read more

LPDDR

Low-power DRAM built for phones and laptops, where battery matters more than raw bandwidth. Each numbered generation moves data faster; LPDDR6 is the newest, and the first phones carrying it shipped in 2026. Read more

M

machine-readable

Written in a structured format a program can process directly, such as rules encoded as data, instead of prose a person has to interpret. Read more

macro placement

Choosing where large circuit blocks sit on a chip. Their positions affect wiring, timing and whether the final layout satisfies manufacturing rules. Read more

macro-average

An average that gives each group or task equal weight, regardless of how many examples it contains. Read more

Magnificent Seven

The seven largest US technology stocks: Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta and Tesla. Together they are roughly a third of the S&P 500's value, so their moves dominate the index. Read more

MAI

The name used for Microsoft AI’s own model family. Its draft conduct code describes intended behavior and training standards for those models. Read more

manager's amendment

A package of changes to a bill offered by its floor managers, often used in the Senate to fold other bills into must-pass legislation like the annual defense bill. Read more

mandatory convertible preferred stock

Preferred shares that must convert into common shares under specified terms, including a conversion date and a formula determining how many common shares the holder receives. Read more

Markdown

A plain-text format for headings, lists, links and other document structure, readable directly or rendered into a formatted page. Read more

market cap

Share price multiplied by shares outstanding: what the public market says a listed company is worth today. Read more

mass gap

The second half of the Yang-Mills Millennium Prize problem: proving that the lightest particle the theory allows has a mass above zero. Existence of the theory is the first half. Read more

materialized view

A saved result of a database query, refreshed as its underlying data changes. It avoids recomputing the full query whenever someone reads it. Read more

MATH-500

A 500-problem subset of the MATH dataset of competition mathematics, used as a standard test of a model's math reasoning. Read more

Mathlib

The shared library of formalized mathematics for the Lean proof assistant, written by a volunteer community. New formal proofs build on its definitions instead of starting from axioms. Read more

MATLAB

A programming and numerical-computing environment used for engineering, scientific analysis and modeling. Some toolboxes require separate license entitlements. Read more

Matryoshka embeddings

Embeddings trained so the first dimensions carry the most meaning, letting a 768-number vector be cut to 128 with modest quality loss. One model then serves both cheap and precise retrieval. Read more

matting

Separating a foreground subject from its background at the level of soft edges like hair, finer than a segmentation outline. Read more

mature node

An older chip manufacturing process, roughly 28 nanometers and larger, used for power, connectivity and car chips. It is cheaper and far less restricted by export rules than leading-edge processes. Read more

MAU

Monthly active users, the count of distinct people who used a product at least once in a month. The standard scale metric for consumer apps, and easy to inflate with a generous definition of used. Read more

MCP

Model Context Protocol. An open standard for connecting AI assistants to tools and data sources, so a capability written once works across different clients. Read more

MCP connector

A connection that lets an AI application access tools or data through the Model Context Protocol. Its permissions determine what the application can read or change. Read more

MDASH

Microsoft's codename for an AI system that searches source code for vulnerabilities and uses additional agents to check whether suspected flaws can be exploited. Read more

merge window

The two weeks at the start of each Linux kernel cycle when new features and large changes are accepted. After it closes, only fixes go in during the weekly release candidates. Read more

metascience

The study of science itself: how research gets funded, published, replicated and evaluated, treated as something that can be measured and improved. Read more

METR

Model Evaluation and Threat Research, a Berkeley nonprofit that tests frontier AI models for dangerous capabilities, often under agreements with the labs that build them. Read more

MFN

Most favored nation. In a startup SAFE, a clause guaranteeing the investor the best terms any later investor negotiates. An uncapped MFN note sets no valuation of its own; it borrows one from the next round. Read more

MGX

Nvidia’s modular server and rack design framework. Hardware makers use its reference designs to integrate processors, networking and other components. Read more

Micro virtual machine

A small isolated computing environment with its own operating-system boundary, used to run code separately from the host and other workloads. Read more

microgrid

A local electricity network with its own generation or storage that can operate separately from the wider grid. Read more

Microsoft Execution Containers

Windows operating-system infrastructure for running persistent agents with execution controls and supervision. A boundary around execution does not establish that the agent makes correct decisions. Read more

microtask

A paid unit of online piecework taking seconds to minutes: label an image, transcribe a receipt, rate an answer. Much of the training data behind early machine learning was assembled this way. Read more

microVM

A stripped-down virtual machine that boots in milliseconds and gives a workload its own kernel. The isolation of a VM at close to the cost of a container, popular for sandboxing AI agents. Read more

middle mile

Freight routes between fixed business endpoints, like a distribution center and a store, as opposed to delivery to somebody's door. Fixed, repeatable routes with no passengers, which is why autonomous trucking starts here. Read more

Millennium Prize Problems

Seven mathematics problems the Clay Mathematics Institute posted $1 million bounties on in 2000. The Poincare conjecture is solved; Navier-Stokes regularity was among the six still open. Read more

MIMIC-IV

A large public dataset of anonymized hospital records from a Boston medical center, widely used to test medical AI. Read more

misalignment

When an AI system pursues goals or behavior that conflict with the intentions or constraints of the people directing it. Read more

MIT license

One of the most permissive open-source licenses: use, modify and sell freely, keep the copyright notice. On model weights it means almost no strings attached. Read more

model adapter

A small set of trainable components added to a model to specialize its behavior without retraining all its original parameters. Read more

model constitution

A written set of principles used to guide a model's behavior or training. It describes intended conduct; it does not guarantee compliance. Read more

model routing

Software that reads each AI request and picks which model answers it, sending easy work to cheap models and hard work to expensive ones. It turns model choice from a policy decision into a per-query price comparison. Read more

Model Spec

OpenAI's public rulebook for how its models should behave, including whose instructions win when they conflict. Read more

model steering

Changing selected internal model signals during use to encourage a behavior without retraining all its weights. Experiments test whether those signals cause an observed capability. Read more

Model weights

The numerical values learned during training that determine how a model processes inputs. Keeping weights frozen means changing the surrounding system without retraining the model. Read more

model-based systems engineering

Designing complex machines such as satellites from a formal model of their parts, requirements and behavior, instead of from documents. Changes are made to the model and checked against it. Read more

ModernBERT

A language model designed to classify and represent text using surrounding context. GitHub adapts it to distinguish likely credentials from harmless strings. Read more

module shadowing

When a local file shares a name with a standard-library module, Python can load the local file instead. An attacker who plants the file turns an innocent import into code execution. Read more

MoE

Mixture of Experts. A model split into specialist sub-networks where only a few run for any given input, so a very large model can be cheap to run. Read more

MOFCOM

China's Ministry of Commerce, which issues the country's export controls, including those on rare earths, as numbered announcements. Read more

monitorability

How well an observer can detect a relevant behavior or influence from a model’s outputs, such as whether its reasoning reveals that a prompt intervention changed its answer. Read more

monorepo

A single code repository holding many of a company's projects at once, common at large tech firms. Reaching it means reaching most of the company's source code in one place. Read more

moratorium

A temporary legal stop on an activity. For data centers it usually means no new permits or approvals until a set date or until new rules are written. Read more

motion capture

Recording how a person or object moves so the movement can be measured or reproduced digitally. Markerless systems estimate movement from images without attaching tracking markers. Read more

Motion prior

A learned representation of plausible movement that helps guide a robot’s actions, reducing the need to learn each motion from scratch. Read more

motion to dismiss

A request asking a court to throw out a lawsuit early because, even if every allegation is true, it does not state a legal claim. Surviving one lets the case move to evidence gathering. Read more

motion to proceed

The Senate vote to begin debating a bill. When it needs cloture, 60 votes open debate, but the bill still faces amendments and a final vote. Read more

MOU

A memorandum of understanding: a signed statement of intent that is not a binding contract. An MOU announces a deal before the terms that would make it one exist. Read more

mRNA

The molecule cells use as working instructions for building proteins. mRNA vaccines deliver those instructions so the body manufactures a target protein and trains the immune system against it. Read more

multi-agent system

Several AI agents working on one task or in one environment, each with its own instructions, passing messages or sharing tools instead of a single agent doing everything. Read more

multi-token prediction

Training a model to predict several upcoming tokens at once instead of only the next one, which can improve quality and speed up generation. Read more

multimodal

A model that handles more than text. Images, audio and video in, and often out, all through one set of weights instead of separate models wired together. Read more

musl

A small C standard library used by lightweight Linux distributions such as Alpine and in many embedded devices, in place of the more common glibc. Read more

MXFP4

A 4-bit number format for storing model weights, with values grouped under shared scale factors. It shrinks a model to a fraction of its 16-bit size. Read more

MXFP8

Low-precision number formats for AI chips that store values in 8 or 4 bits and share one scaling factor across a small block, so training runs faster with little loss of accuracy. Read more

N

n-day

A security flaw that has already been publicly fixed but is still exploitable on machines that have not installed the fix. The counterpart to a zero-day, which nobody has patched yet. Read more

n-gram

A run of n consecutive words or tokens. Storing statistics for common n-grams lets a model look up frequent patterns cheaply instead of computing them, an old technique now returning inside frontier architectures. Read more

NAND flash

The memory chips inside SSDs and USB drives, storing data by trapping electric charge in cells rather than spinning a disk. Cheaper and slower than the DRAM used for a computer's working memory. Read more

NASK

Poland’s national research institute for information and communications technologies, with responsibilities in cybersecurity and AI safety. Read more

Navier-Stokes

Equations describing fluid motion, including viscosity. The existence and smoothness problem asks whether smooth flows always remain well behaved under specified conditions. Read more

NCMEC

The US clearinghouse for reports of child sexual abuse material. Online platforms are legally required to report such material to it through the CyberTipline. Read more

NCSC

The UK's national cybersecurity agency, part of the GCHQ intelligence service, which handles major incidents and publishes threat assessments. Read more

NDAA

The annual US law that sets Pentagon policy and spending limits. Because it must pass every year, lawmakers attach unrelated measures to it, such as chip export rules. Read more

NDRC

China's state economic planning agency, which drafts five-year plans and approves major investment and industrial policy, including on AI. Read more

near-packaged optics

Optical connections placed close to a computing chip’s package, shortening the electrical path before data travels as light. The aim is greater bandwidth with less connection power. Read more

nearline storage

Data-center hard drives for data used too often for tape but not often enough for fast SSDs. The cheap bulk tier where training corpora, logs and archives sit, and the storage segment AI demand has pushed into shortage. Read more

neocloud

A cloud provider built around renting GPUs for AI rather than general-purpose computing. CoreWeave, Nebius and Lambda are the type specimens; most lease chips from Nvidia and sell capacity on to the labs. Read more

NEPA

The National Environmental Policy Act, the 1970 US law requiring federal agencies to study a project's environmental effects before approving it. Its reviews can take years and invite lawsuits. Read more

net proceeds

Money an offering leaves for the company after transaction expenses. Gross proceeds are the amount raised before those deductions. Read more

net revenue

Revenue after paying a partner its share of a sale. A company that books cloud-marketplace sales net reports less than one that books the full sale gross, so the two cannot be compared directly. Read more

neural operator

A neural network that learns mappings between functions rather than fixed-size inputs, so one trained model can simulate physics like fluid flow at any resolution. The architecture behind several AI weather and engineering models. Read more

neural rendering

Using trained neural networks to generate or reconstruct parts of an image, often reducing the work required by conventional graphics rendering. Read more

NGC

Nvidia’s catalog for distributing software containers, models and other computing resources. Packaged products can have terms separate from their source code. Read more

NIPRNet

The Defense Department's network for unclassified but sensitive information, kept separate from the public internet and from its classified networks. Read more

NIST

The US National Institute of Standards and Technology, a federal agency that develops measurement methods and technical standards, including work on AI risk. Read more

Nix

A package manager and build system that records a program’s dependencies precisely, helping developers reproduce the same software environment and keep different versions apart. Read more

noise-space learning

A way to adapt a generative robot policy by learning the noise inputs that guide its actions while keeping the underlying pretrained model fixed. Read more

non-cash charge

An accounting expense that reduces reported profit without cash leaving the company, such as depreciation or remeasuring liabilities that convert into shares. It can turn a moderate operating loss into a spectacular net one. Read more

non-compete clause

A contract term barring an employee from joining or starting a rival for a period after leaving. California refuses to enforce them, which startup advocates credit for Silicon Valley's job mobility. Read more

non-disparagement clause

A contract term barring someone from saying negative things about a company, often for years after they leave. AI staff have objected because it can stop them raising safety concerns in public. Read more

Non-exclusive license

Permission to use technology that leaves its owner free to license the same technology to others. Read more

non-GAAP

Figures a company reports outside standard accounting rules, usually leaving out stock pay and one-off costs. They tend to look better than the official numbers. Read more

non-human identity

A login held by software instead of a person, such as an API key or service account. AI agents multiply them, and each one is a credential to steal. Read more

non-inferiority

A study test asking whether a method performs no worse than a comparison method by more than a predefined acceptable margin. Read more

non-sofic

A group that cannot be approximated by finite ones. Whether any exists was asked by Gromov in 1999; producing one is the kind of result that had resisted every human attempt for decades. Read more

nonbinding resolution

A measure in which a chamber of Congress states a position without changing any law. It signals intent and puts members on record. Read more

noncash charge

An expense or accounting adjustment that reduces reported profit without an equivalent cash payment in that reporting period. Read more

nonfarm payrolls

The monthly count of US jobs added outside farming, from the Bureau of Labor Statistics survey of employers. The single number markets trade hardest, released the first Friday of each month. Read more

nonvoting stock

Shares that carry an economic interest in a company without ordinary shareholder voting rights. They let a company raise capital without distributing equivalent control. Read more

notice of appeal

The short filing that starts an appeal and preserves the right to one. The legal arguments come later, in briefs. Read more

npm

A widely used registry and package manager for JavaScript software. Developers download dependencies from it and publish reusable packages to it. Read more

NPU

Neural processing unit: a chip built only for neural-network arithmetic, usually for running models rather than training them. More power-efficient than a GPU at the jobs it is designed for, and less flexible. Read more

nuclear command and control

The systems and procedures by which a country's leaders authorize, communicate and carry out decisions to use nuclear weapons. Keeping AI out of it means a human makes the launch decision. Read more

nuclear uprate

An upgrade that increases the permitted power output of an existing nuclear reactor, usually through equipment or operating changes. Read more

NVFP4

Nvidia's 4-bit floating-point format for model weights, supported in hardware on its Blackwell and later chips. Half the memory of FP8 again, with a per-block scale factor to limit the loss of precision. Read more

NVL72

Nvidia's rack-scale format: 72 GPUs wired by NVLink into what software sees as one enormous accelerator. Big orders are now counted in racks rather than chips. Read more

NVLink

Nvidia's high-speed interconnect linking chips inside a server rack far faster than standard networking. NVLink Fusion licenses that fabric to other chipmakers, so custom silicon can slot into Nvidia systems. Read more

O

OAuth

The standard that lets an app act on a user's account at another service through a token, without ever holding the password. Stolen tokens or secrets let an attacker do whatever the app was allowed to do. Read more

observability

Using logs, measurements and recorded interactions to understand what a running system did and why it failed. The records may contain sensitive user data. Read more

OCR

Optical character recognition: turning images of text, such as scanned pages, into text a computer can edit and search. Read more

off-balance-sheet

An arrangement where the assets and debt sit in another entity, so they never appear in a company's own accounts. Legal, common, and the reason a balance sheet can understate what a company has committed to. Read more

offshore holding company

A parent company registered abroad, often in the Cayman Islands, that Chinese startups used so foreign investors could own them and list overseas. Beijing's rules now push some to unwind it. Read more

omnimodal

A model that takes text, images, audio and video in one request rather than routing each to a separate system. Vendor shorthand extending multimodal to native audio-video understanding. Read more

Ontology

A shared description of concepts and their relationships. In business software, it defines how records such as customers, products and orders fit together. Read more

opaque serial depth

A proposed measure of how much sequential computation an AI performs between outputs people can interpret. It estimates hidden reasoning capacity, not whether a system is safe. Read more

Open Compute Project

An industry community that publishes shared designs and specifications for data center hardware, so suppliers can build compatible systems. Read more

open interest

The total outstanding derivative contracts that have not been closed. It measures existing positions, while trading volume measures transactions over a period. Read more

Open VSX

An open registry of extensions for VS Code-style editors, run by the Eclipse Foundation. Editors that cannot use Microsoft's own marketplace, Cursor among them, install extensions from it. Read more

open-weight

A model whose trained parameters are published, so anyone can download and run it themselves. Not the same as open source: the training data and code usually stay private. Read more

OpenRouter

A service that resells access to hundreds of AI models from different providers through one API and one bill, so a program can switch models by changing a name. Read more

operating income

Revenue minus the costs of running the business, before interest, taxes and gains or losses on investments. Dividing it by revenue gives the operating margin. Read more

Opportunity Zone

A US program that cuts capital gains taxes for investments in designated low-income areas. Several data center projects have used it. Read more

optical circuit switch

A network switch that redirects light between fibers without converting it to electrical signals, saving the power and delay of the conversion. Google runs them inside its own AI clusters. Read more

optical interconnect

A connection that carries data as light between computing components. It can extend reach and reduce electrical losses, but requires hardware to convert and manage the signals. Read more

optical transceiver

The plug-in module that converts electrical signals to light and back at each end of a fiber link. Data centers buy them by the million, and Chinese vendors lead the market. Read more

OSWorld

A family of evaluations that asks AI agents to complete tasks on a computer. Scores depend on the version and on which tools the agent is allowed to use. Read more

out-of-band release

A software update issued outside the normal release schedule, often to fix a security vulnerability that cannot wait. Read more

out-of-distribution

Inputs unlike anything in a model's training data. Models often stay confident while failing on them, which is why a record El Niño tests AI weather forecasters. Read more

Outcome verifier

A separate checker that judges whether an action achieved the intended result. Its usefulness depends on which failures it can detect. Read more

overallotment option

The right an IPO's underwriters have to sell extra shares beyond the original offering if demand runs high, raising the total amount the company collects. It is why a listing's final size often exceeds the number first announced. Read more

oversubscribed

An offering is oversubscribed when investors ask for more shares than are for sale; the multiple measures demand. At the extreme, allocations are decided by lottery. Read more

overweight

An analyst rating meaning a stock should make up more of a portfolio than its share of the benchmark index. Roughly equivalent to buy. Read more

P

P/E

Price-to-earnings ratio: a company's share price divided by its yearly earnings per share. Listing at 219 times earnings means a buyer pays 219 years of current profit up front, a bet that profit grows into the price. Read more

par

A bond's face value, usually quoted as 100 cents on the dollar. A bond trading at 79.5 cents is priced about 20% below par, which signals that buyers doubt they will be repaid in full. Read more

paralinguistics

The part of speech that is not the words: tone, pace, stress, hesitation. Transcription drops it, which is why two identical transcripts can carry opposite meanings. Read more

parameter-efficient fine-tuning

Adapting a large model by training a small add-on, like a LoRA adapter, instead of all its weights. It makes fine-tuning affordable when data or compute is scarce. Read more

parameters

How many individual numbers a model has learned. A 30B model has thirty billion. It is a rough size measure and a poor quality measure, since a well-trained small model routinely beats a badly trained large one. Read more

parent guarantee

A promise by a parent company to pay if its subsidiary cannot. Landlords ask for one when the tenant signing the lease is a smaller unit of a bigger, better-rated company. Read more

pass@k

The share of problems a model solves at least once in k attempts. Pass@1 measures a single try; comparing the two shows whether retrying helps. Read more

password hash

A scrambled form of a password that a site stores in place of the password itself. A stolen hash can still be cracked if the password is weak. Read more

patch gap

The time between a software maker publishing a security fix and users actually installing it. Attackers can study the published fix to find the bug and exploit anyone still unpatched. Read more

Pax Silica

A US-led declaration in which countries commit to cooperate on AI chips, critical minerals and supply chains outside China's reach. It is not a binding treaty. Read more

penetration testing

Authorized hacking that probes systems for exploitable weaknesses before attackers find them. The same tooling works without authorization, which is why its automation cuts both ways. Read more

PERM

The Labor Department certification that starts most employer-sponsored green cards: the employer must show no qualified US worker took the job. Suspending a company freezes its new green-card pipeline, not its existing visas. Read more

perpetual futures

Derivative contracts with no fixed expiry. Traders gain exposure to a reference price without owning the underlying asset; the contract price can differ from an actual share transaction. Read more

personal access token

A credential string that stands in for a password when software talks to a service like GitHub. Whoever holds the token holds its permissions, which is why one leaked in a public file is a full breach. Read more

perturbation data

Measurements of how cells change when a gene is silenced or a chemical is added, used to train models that predict what a drug will do to a cell. Read more

Phase 1

An early human study that primarily evaluates a treatment’s safety and dosing. Starting one does not establish that the treatment works. Read more

phishing

Impersonating a trusted person or service to trick someone into disclosing credentials or granting access. Some attacks relay a real sign-in and steal the resulting authenticated session. Read more

photomask

A patterned template used to project circuit designs onto a semiconductor wafer during chip manufacturing. Read more

photonic processor

A computing device that uses light for some operations. Its useful speed and energy cost also depend on memory, electronic control and conversion between light and electrical signals. Read more

PHP

A programming language widely used to run websites. A flaw in its runtime can affect many different applications built on it. Read more

physics engine

Software that simulates how objects move, collide and push on each other. Robot policies are trained in thousands of simulated copies before touching a real robot, so its speed sets training cost. Read more

physiological

Relating to how a living body functions, such as breathing, blood flow or heart motion. Read more

PIPE

A sale of shares to selected investors at a fixed price alongside a listing or merger, often used to top up a SPAC deal when public shareholders redeem their money. Read more

PJM

The regional grid operator coordinating wholesale electricity and transmission across parts of the eastern and midwestern United States. Read more

PKCE

A check in OAuth that ties an authorization code to the app that requested it, so an intercepted code is useless alone. A leaked PKCE key removes that protection. Read more

Playwright

Software for controlling web browsers through code, commonly used to test websites and automate browser tasks. Read more

PNG

A common image-file format that supports transparency, letting a cutout or separate layer sit over another image. Read more

polysemanticity

The appearance that one artificial neuron responds to several unrelated concepts. Researchers investigate whether those responses reflect shared structure or genuinely different functions. Read more

post-money valuation

A startup's value counting the new investment just raised. A $2M round at a $10M post-money means investors own 20%; the pre-money was $8M. Read more

post-quantum

Cryptography designed to resist attacks from future quantum computers, including attacks against methods that protect current internet traffic. Read more

post-training

Everything done to a base model after its main pretraining run: supervised fine-tuning, reinforcement learning and reasoning tuning. Labs now get large capability gains here without changing the base model at all. Read more

PostgreSQL

An open-source relational database that stores structured records and supports transactions. It can also coordinate durable job queues. Read more

power purchase agreement

A contract to buy electricity over an agreed period. Long contracts can give a generator predictable revenue to finance construction or keep a plant operating. Read more

pre-money valuation

A startup's value before a new round of funding is added in. A $20M raise at a $3.8B pre-money valuation becomes a $3.82B company once the round closes. Read more

pre-seed

The earliest outside money a startup raises, usually before it has a product. Sums run from a few hundred thousand to a few million. Read more

precision

The share of a system’s reported findings that are correct. A reviewer can achieve high precision by reporting only obvious defects while missing many others. Read more

preclinical

Research conducted before testing a treatment in people. Laboratory results can support further development without establishing that a treatment is safe or effective for patients. Read more

preemption

A federal law overriding state ones on the same subject. In US AI policy it is the central fight: whether Washington sets a single standard or fifty states each set their own. Read more

preferred stock

Shares that get paid before common stock in a sale or liquidation, usually with fewer voting rights. Most private startup investors hold preferred. Read more

prefill

The first stage of running a language model: reading the whole prompt at once, which is compute-heavy. The second, decode, writes the answer a token at a time and is limited by memory speed, so some systems run the two on different chips.

prefix caching

Reusing computation for the shared opening of several prompts. Prefix-aware routing sends those requests to the server that already holds the reusable state. Read more

premarket

Trading in US stocks before the exchange's 9:30am New York open, on thin volume. Premarket moves show direction, and they often shrink or reverse once full trading starts. Read more

Preparedness Framework

OpenAI's internal rulebook for measuring dangerous model capabilities before release, with thresholds like High and Critical that trigger stronger safeguards. Anthropic's equivalent is its Responsible Scaling Policy. Read more

preprint

A research paper shared publicly before formal peer review. Its methods and results are available to inspect, but publication alone does not establish independent validation. Read more

preservation letter

A legal demand to keep every document related to a matter because litigation or an investigation may follow. Deleting records after receiving one can itself become the offense. Read more

Presidential Records Act

The 1978 law that makes a president's official records public property, to be preserved and handed to the National Archives. Read more

press publishers' right

An EU copyright rule from 2019 that gives news publishers control over online reuse of their articles. Reporting a bare fact stays free. Read more

pretraining

The initial training stage in which a model learns general patterns from large datasets, before later training adapts it to instructions or specific tasks. Read more

price target

An analyst's published estimate of where a stock trades within about a year. Targets move markets less by being right than by showing which way professional conviction is shifting. Read more

price-to-sales

Market value divided by a year of revenue, used to price companies without profits. A ratio above 100 means investors are paying today for a century of current sales. Read more

primary capital

Money a company receives by issuing new shares. A sale of existing shares instead pays the shareholders selling them.

prior authorization

A US insurer's requirement that a doctor get approval before a treatment, test or drug is covered. Physicians file about 40 a week on AMA figures, which is why it is an early target for AI.

private credit

Loans negotiated outside public bond markets, usually from investment funds. Their terms can be tailored to a borrower’s revenue and assets. Read more

private placement

A sale of shares directly to chosen investors rather than on the open market. In an IPO it lets a strategic partner buy a stake at the offer price without competing for allocation. Read more

Private Safety Processing

OpenAI's automated safety review of encrypted records in customer-controlled storage. Review runs in a protected computing environment designed to prevent personnel from reading the content. Read more

procedural graph

A map of steps and their relationships that guides an agent through a task. A system can revise the map using evidence from successful and failed attempts. Read more

process mining

Using records of business activity to reconstruct how work actually flows, exposing delays, repeated steps and departures from the intended process.

process node

The generation of chipmaking technology a processor is built on, named in nanometers. Smaller numbers pack more transistors into the same area; the labels are marketing shorthand rather than literal measurements. Read more

process tree

A running program and the child processes it starts, including programs started by those children. Stopping the original process does not always stop all its descendants. Read more

production yield

The share of chips made that actually work and can be sold. Low yields mean a factory wastes wafers and ships less than its capacity. Read more

programmable logic controller

An industrial computer that operates physical equipment, valves, pumps, motors, on a factory floor or at a water or power plant. Decades old, rarely patched, and increasingly reachable from the open internet. Read more

prokaryote

A single-celled organism with no nucleus, such as a bacterium. Human, animal and plant cells are a different group, called eukaryotes. Read more

prompt caching

Storing a prompt's already-processed context so repeat requests reuse it at a steep discount instead of paying full input price. Agent loops re-read the same context constantly, so cache pricing often matters more than list pricing. Read more

prompt injection

An attack that hides instructions in content a model reads, so the model follows the attacker instead of its operator. The AI equivalent of SQL injection, and still unsolved. Read more

proof assistant

Software that checks every step of a mathematical proof, so a proof it accepts is correct. Lean is the most widely used one. Read more

prospectus

The document a company publishes before selling shares, setting out its finances and risks. In Hong Kong and Europe it is the equivalent of the US S-1. Read more

Protease

An enzyme that cuts proteins. A designed protease must be tested on physical samples to establish that its intended reaction actually occurs. Read more

provenance

The record of where data or an artifact came from and how it was produced. It helps someone trace a result back to its sources and processing steps. Read more

Psychometrics

Methods for testing whether a score measures the ability or behavior it claims to measure. In AI, this can reveal whether a benchmark mixes several different skills. Read more

public nuisance

An old legal claim against conduct that harms the general public's health, safety or use of shared resources. It was used against tobacco and opioid makers and lets a court order the conduct to stop. Read more

Public Service Commission

The state body that regulates utilities in many US states, setting electricity rates and deciding who pays for grid upgrades. Read more

public-benefit corporation

A for-profit company legally required to weigh a stated public mission alongside shareholder returns. Anthropic is one, incorporated in Delaware. Read more

PubMed

A searchable index of biomedical publications. A PMID identifies a record in that index; it does not establish that the paper supports a particular claim. Read more

pull request

A proposed code change submitted for review before it is merged into a project. Its existence does not mean the change is available in a stable release. Read more

pull_request_target

A GitHub Actions event that runs in the base repository’s context. Executing untrusted contributor code in such a workflow can expose privileged credentials. Read more

purchase obligations

Contracts a company has signed to buy goods or services in future, such as chips from suppliers, which it must pay for whether or not demand holds up. Read more

PURPA

The Public Utility Regulatory Policies Act of 1978, a US federal law that sets standards state utility regulators must formally consider. Amending it is how Congress steers electricity policy without commanding the states. Read more

PutnamBench

A benchmark of 672 problems from the Putnam undergraduate math competition, written so a proof assistant can check a model's proofs mechanically. Read more

Pwn2Own

A hacking contest where researchers are paid for working attacks on fully patched products, which vendors then get a fixed window to repair before details are published. Read more

Pyodide

A version of Python built to run through WebAssembly, including in browsers and supported server environments. Native Python packages may need special support. Read more

Python

A general-purpose programming language used for automation, data analysis and AI software. Its readable syntax and large library ecosystem make it common for quickly building and connecting tools. Read more

PyTorch

An open-source software framework used to build, train and run machine-learning models. Read more

Q

QLC

A flash-memory cell that stores four bits instead of one, packing more capacity onto the same chip at the cost of speed and durability. The current way to make NAND flash cheaper per terabyte. Read more

quadruple patterning

Printing one chip layer in four aligned lithography passes to reach features finer than the tool could draw in one. It works without EUV but adds steps per wafer, so cost and yield suffer. Read more

quantization

Storing a model's numbers at lower precision so it fits in less memory and runs faster. INT8 and BF16 name how many bits each number gets. Some accuracy is lost, usually less than you would expect. Read more

qubit

A unit of quantum information. Quantum circuits manipulate qubits through operations called gates; noise and errors limit how reliably they can compute. Read more

Query fan-out

How an AI search system answers one question by splitting it into many smaller searches, running them at once and combining what comes back. It is why an AI Mode answer can cite pages a single search would never have surfaced. Read more

R

rack-scale

Selling a whole data center rack of chips, networking and cooling as one machine, in place of separate servers. Nvidia's NVL systems and AMD's Helios are built this way. Read more

RAISE Act

New York's frontier AI safety law. From January 2027, developers with over $500 million in revenue must publish safety frameworks, report critical incidents within 72 hours and face fines up to $3 million. Read more

randomized controlled trial

An experiment where chance decides who gets which treatment, so any difference in outcomes can be traced to the treatment and not to who chose it. Read more

ransomware

Malware that encrypts or steals a victim's data and demands payment to restore it. Run by criminal gangs as a business, complete with affiliates, negotiation portals and customer service. Read more

rare earths

Seventeen metals used in the magnets inside motors, drives and chips' supply chains. China mines and refines most of the world's supply, which makes export limits a lever. Read more

RDIMM

A registered memory module, the standard form DRAM takes in servers. A buffer chip between memory and processor lets a server run far more memory per socket than desktop modules allow. Read more

React Native

A framework for building mobile apps with React and JavaScript while using the phone platform’s native interface components. It lets developers share substantial code between Android and iOS. Read more

reasoning model

A model trained to work through a problem in steps before answering, spending more compute per question. Better at math and code, slower and dearer at everything. Read more

recall

The share of relevant cases a system finds. In code review, low recall means real defects escape detection even if the comments it does produce are accurate. Read more

recall query

An NHTSA investigation into whether a vehicle recall actually fixed the defect it was issued for. It can end with a new recall. Read more

recursive self-improvement

The hypothesized loop where an AI system improves its own ability to improve, compounding capability faster than human oversight can follow. Central to most catastrophic-risk arguments. Read more

red team

A group paid to attack a system before its users do. For models it means deliberately trying to make one produce the output it was built to refuse. Read more

redemption

Returning shares to an issuer in exchange for cash. In a SPAC deal, shareholders withdrawing their cash can reduce the money available at the merger. Read more

regression

A change that makes a previously working behavior worse. Regression testing checks that a repair or new feature has not broken existing behavior. Read more

reinforcement learning

Training a model by rewarding good outputs and penalizing bad ones, so it learns from feedback on its own attempts rather than only imitating examples. The stage where much of a frontier model's reasoning is now shaped. Read more

release candidate

A software version offered for testing before a final release. Further defects may lead to another candidate instead of immediate general release. Read more

remaining performance obligations

Contracted revenue a company has signed but not yet delivered. A cloud provider's RPO shows demand years ahead of the income statement, and it can be renegotiated or canceled before any of it becomes revenue. Read more

remediation

Fixing a security weakness and deploying the repair. Finding a vulnerability alone does not remove the risk. Read more

remote code execution

A vulnerability that lets an attacker run their own code on someone else's machine, the most serious class of software bug. Read more

remote monitoring and management

Software that IT service providers use to watch, patch and control their clients' computers from one console. A flaw in it can hand an attacker every client network at once. Read more

replay

Re-running recorded work with saved inputs or responses. Replacing a model’s answers alone does not reproduce the behavior of the external tools it used. Read more

replication data

The data and code a study's authors publish so others can rerun the analysis and check that the reported numbers come out the same. Read more

reranker

A model that reorders candidates returned by an initial search, trying to put the most relevant results first. It does not necessarily search the underlying collection itself. Read more

research closure

Whether an investigation supplies the evidence and checks needed to answer its research question, beyond completing individual calculations or intermediate tasks. Read more

residual value

The value an asset is expected to retain after use or at the end of a lease. A promise to support that value exposes the guarantor to depreciation. Read more

residual value guarantee

A promise to cover the gap if leased equipment is worth less than expected when the lease ends. It lets a chipmaker backstop a customer's debt without booking the risk as a liability. Read more

Responsible Scaling Policy

Anthropic's published rulebook tying how far it trains and deploys models to measured capability thresholds, with stronger safeguards required, or work paused, as each threshold is crossed. Read more

retrieval-augmented generation

A setup in which a model looks up documents at the moment it answers and works from them, instead of relying only on what it learned in training. Read more

reverse shell

A connection a compromised machine opens outward to an attacker, giving them a command line inside the network. Firewalls that block incoming traffic often allow it, because the victim dialed out. Read more

reverse transcriptase

An enzyme that copies RNA into DNA. Finding one in a genome does not by itself establish what biological job its surrounding system performs. Read more

reverse tunnel

A connection a compromised machine opens outward to an attacker's server, so the attacker can reach back in through a firewall that would block a direct connection. Read more

revolving credit facility

A pre-arranged bank loan a company can draw down, repay and redraw as needed, up to a limit. A standby source of cash rather than a one-time loan, often lined up before a public listing. Read more

reward hacking

When an AI system exploits flaws in how its goal is scored instead of doing the intended task, like tampering with the grader rather than solving the problem. A central failure mode in training agents with reinforcement learning. Read more

Riemann zeta

A function at the center of how prime numbers are distributed. Where its value hits zero is the subject of the Riemann hypothesis, unsolved since 1859 and the most famous open problem in mathematics. Read more

RISC-V

An open instruction-set architecture: the commands a processor understands. Companies can use it to design chips without making their own chip designs public. Read more

risk factor

The section of an IPO prospectus where a company lists what could plausibly hurt the business, written partly to shield it from investor lawsuits. What a company chooses to name there is a map of what its bankers fear. Read more

RL environment

A simulated task, such as a codebase, a spreadsheet or a web app, with a way of scoring the result, in which a model practices and is rewarded for getting it right during reinforcement learning. Read more

RLHF

Training that has people rate a model's outputs and steers the model toward what raters preferred. The rating is paid piecework at industrial scale, increasingly done by the professionals whose judgment the model is learning. Read more

roadshow

The tour a company makes to pitch large investors before listing. It is where the story gets tested, and where the price range usually moves. Read more

robotaxi

A ride-hail vehicle that drives itself with no human behind the wheel, dispatched and paid like a taxi. Waymo, Zoox, Tesla and several Chinese operators run competing fleets. Read more

robotic process automation

Software that performs repetitive steps in business applications, often through their existing interfaces when a direct API is unavailable. Read more

robots.txt

A file websites publish telling automated crawlers which pages not to fetch. It is a request, not an enforcement mechanism: compliance is voluntary, which is why AI scrapers can ignore it. Read more

RoboTwin

A simulated robotics benchmark for testing manipulation policies on tasks involving objects and robot arms. Performance in its tasks is distinct from reliability in a deployed workplace. Read more

robovan

A driverless delivery vehicle sized between a sidewalk robot and a truck, built to carry parcels on public roads rather than passengers. Read more

rollback

Restoring a system to an earlier recorded state after a failure. It can undo reversible software changes, but cannot retract an already delivered message or physical action. Read more

romanization

Writing a language with Latin letters instead of its customary script. It can produce spellings that differ across speakers and regions. Read more

ROS

An open-source framework of libraries and tools for building robot software, widely used across research and industry despite the name; it is not an operating system in the traditional sense. Read more

RSA

A public-key encryption and signature system from 1977 whose security rests on how hard it is to factor large numbers. Keys of 1024 bits are considered legacy; 2048 bits is the usual minimum now. Read more

RSS

A web standard that publishes a machine-readable feed of a site's new posts, so feed readers and other software can follow updates without visiting the page. Much of the open web has syndicated through it since the early 2000s. Read more

RTK

A command-line utility that reduces terminal output passed to coding agents. Smaller output can reduce context use, but the effect on the final bill depends on the agent’s subsequent work. Read more

RTL

Register-transfer level: the source code of a chip, written in languages like Verilog, describing how data moves between registers on each clock cycle. What designers write before silicon or an FPGA runs it. Read more

Rubin

Nvidia's data-center chip generation after Blackwell. Vera is the CPU and Rubin the GPU, sold together in rack-scale systems. Read more

rubric

An explicit set of criteria for grading an answer or task, defining what the evaluator should check rather than leaving quality to an overall impression. Read more

Rule of 40

A rough test for software companies: revenue growth rate plus profit margin, both in percent, should add up to at least 40. Palantir reports its own version each quarter. Read more

run-rate

Current revenue projected forward a full year. A company earning $4B this month has a $48B run-rate, which flatters fast growth and is why it gets quoted so often. Read more

runoff

A second round of voting between the top two finishers when no candidate wins a first-round majority. Brazil's presidential elections use one. Read more

runtime

The period when a program is executing. A choice known only at runtime depends on live inputs rather than information fixed when the program was written. Read more

Rust

A programming language designed for fast systems software, with rules that prevent many memory-access errors before a program runs. Read more

S

S-1

The registration document a US company files with the SEC before going public. A confidential S-1 is filed privately first, so the company can start the process without publishing its numbers. Read more

S-1/A

An amendment to the registration statement a company files with the SEC before an IPO. Companies file several, adding fresh quarterly results and, close to launch, the price range and share count. Read more

SAFE

Simple Agreement for Future Equity. An investment that converts into shares later, at a price set by a future round, so nobody has to agree what the company is worth today. Read more

safety case

An argument, supported by evidence, that a system is acceptably safe in a specified operating environment. It must address the conditions in which that system will actually be used. Read more

sandbagging

A model doing worse on a task than it is able to, on purpose or by learned habit, without saying so. It matters most on evaluations meant to measure what the model can do. Read more

sandbox

An isolated environment where code runs without touching anything outside it. A sandbox escape means the isolation failed and the code reached the real system. Read more

SCADA

Supervisory Control and Data Acquisition: software that monitors industrial equipment, records its state and lets operators send control commands. Read more

scaffold

The prompts, tools and control software around a model that organize how it attempts a task. A score for the assembled system does not isolate the model’s contribution. Read more

scattering amplitude

A mathematical quantity used to calculate how likely particles are to interact in a particular way. More detailed calculations require increasingly difficult corrections. Read more

schema

A specification for the fields and structure of data. For model output, it can require named fields in a particular order, but cannot guarantee their contents are correct. Read more

SDK

A package of code and utilities that helps developers use a service or platform from their own programs. Read more

SEC

The Securities and Exchange Commission, the US markets regulator. Companies going public file with it, and those filings carry legal liability for what they say. Read more

second lien

A claim on collateral that ranks behind the first lien. Its lender bears more risk if the collateral cannot cover both obligations. Read more

Second Request

A formal US antitrust demand for more information before a merger can close, issued when regulators suspect a deal might harm competition. Receiving one delays a deal and signals real scrutiny, not a routine filing. Read more

secondaries

Trading shares of a private company between existing holders, with no new shares issued. A useful signal because employees and early investors sell what they privately doubt. Read more

secondary sale

A sale of existing shares by employees or early investors, with the money going to the sellers rather than the company. It sets a valuation without raising any new capital. Read more

Section 232

A US trade law that lets the president restrict imports the Commerce Department finds threaten national security, by tariff, quota or negotiated deal. Read more

Section 301

The US trade law that lets the president impose tariffs over unfair foreign practices. It is the basis for most tariffs on Chinese goods; exclusions exempt listed products. Read more

securities class action

A lawsuit brought by one shareholder on behalf of everyone who bought a stock during a set period, claiming the company misled the market. A judge must approve the class before it proceeds. Read more

securitization

Bundling loans or lease payments into a security that investors buy, so the cash flows can be rated and sold apart from the company that created them. Read more

sell-side

The sell-side is banks and brokers that publish research and underwrite deals; the buy-side is the funds that actually invest. Their forecasts for the same company often differ. Read more

semantic versioning

A version-numbering rule: MAJOR.MINOR.PATCH, where only a major bump may break existing users, so a developer can tell from the number whether an upgrade is safe. Read more

senior secured notes

Bonds first in line for repayment and backed by specific collateral. If the borrower fails, these holders are paid from the pledged assets before anyone else sees a dollar. Read more

Sepolia

An Ethereum network used to test software. Its test tokens do not represent ordinary main-network Ether, but using another person's account still requires permission. Read more

serialized execution

Running changes to shared state in a controlled sequence so concurrent requests do not overwrite or conflict with each other. Read more

Series A

Lettered venture funding rounds, in order: A is usually the first large institutional round, with B, C and onward marking later, bigger raises. The letter signals stage, not size, and late letters at small amounts often signal trouble. Read more

settlement layer

The system where transactions finally clear and get recorded, whatever route they took to arrive. Whoever runs settlement gets paid on every trade regardless of which side wins it. Read more

SGLang

A widely used open-source serving engine for running language models in production, comparable to vLLM. Read more

SHA pinning

Locking software to one exact version by its cryptographic hash, so an upstream change cannot silently alter what runs. A pinning bypass lets an attacker swap the code anyway. Read more

shader core

A GPU processing unit that runs small programs used to draw images and perform parallel calculations. Read more

shareholder derivative suit

A lawsuit shareholders bring on the company's behalf against its own executives or directors, claiming they harmed it. Any money recovered goes to the company. Read more

Sherman Act

The 1890 US antitrust law. Section 1 bans agreements between competitors that restrain trade, such as fixing prices or cutting supply together. Read more

short drama

Serialized fiction released in one- to two-minute vertical episodes for phones, usually behind a paywall after the first few. China built it into a large industry. Read more

Show HN

A category on Hacker News where people post things they built. One of the few remaining places a project with no marketing budget gets found. Read more

show-cause order

A regulator's demand that a company either justify its current practice or change it, with a deadline and the burden of proof on the company. FERC used one to force US grid operators to defend their data-center hookup rules. Read more

side effect

A change an operation makes outside its returned answer, such as writing a file or sending a message. Retrying the operation can repeat that change. Read more

sidecar

A South Korean market circuit-breaker that briefly suspends program trading, halting bulk buy or sell orders for five minutes when a stock-index future swings sharply, to slow a cascade. Read more

SIEM

Security information and event management: the software a security team uses to collect and search logs from across a company. Read more

silicon interposer

A silicon layer that connects components inside a chip package, such as a processor and nearby memory. Read more

silicon photonics

Chips that guide and process light, often alongside electronic components, to move data through computing systems. Read more

single sign-on

One login granting access to many separate applications through a central identity provider. Convenient, and a single point of failure: subvert the sign-on step and every connected app opens. Read more

SOFR

The benchmark US dollar borrowing rate that replaced LIBOR. Loans are quoted as a spread over it, so "5.5 points over SOFR" means the benchmark plus 5.5%. Read more

solid-state transformer

A transformer built from power electronics instead of copper coils and iron. It can turn high-voltage grid power directly into the DC a data center rack uses, in less space and with fewer conversion steps. Read more

Source control

Software that records changes to files and lets people compare, combine or recover earlier versions. Git and Scion are examples. Read more

sovereign AI

A government building or funding its own AI models, compute and data infrastructure so the capability sits under national control rather than a foreign vendor. Saudi Arabia's HUMAIN and the EU's various compute funds are the pattern. Read more

sovereign wealth fund

A state-owned investment fund, usually built from oil revenue or trade surpluses. Saudi Arabia's Public Investment Fund and Abu Dhabi's MGX are both active AI investors. Read more

SPAC

A special-purpose acquisition company: a listed shell that merges a private company onto the stock market without a traditional IPO. Faster, with lighter scrutiny, and the track record shows what that trade costs. Read more

sparse activation

Using only part of a model’s parameters for each input. This can reduce computation while the full set of parameters still needs storage. Read more

sparse attention

A method that lets a model consult selected parts of its input instead of comparing every part with every other part, reducing computation. Read more

sparse autoencoder

A model used to break another model’s internal activity into a smaller set of active features that researchers can inspect. Its features are an imperfect view of the underlying computation. Read more

sparse linear system

A set of equations in which most coefficients are zero. Specialized solvers avoid storing and calculating all those empty entries, saving memory and computation. Read more

spatial mask

A representation marking where an object or region sits in an image. A sequence of masks can track how its position changes. Read more

speaker diarization

Splitting an audio recording by who is speaking when, so a transcript reads as labeled turns instead of one undifferentiated stream. Read more

special-purpose vehicle

A separate legal entity created to hold one set of assets and debts, keeping them off the sponsor's balance sheet. AI chip financing increasingly runs through SPVs that buy hardware and lease it back. Read more

speculative decoding

A serving trick where a small draft model proposes several tokens and the big model verifies them in one pass. Same output, delivered faster and cheaper. Read more

speech recognition

Software that turns spoken audio into text. Accuracy is usually reported as word error rate, the share of words it gets wrong. Read more

spin-off

Turning a division into a separate company with its own shares, often listed on an exchange, while the parent usually keeps a stake or control. Read more

spot market

Buying something at today's price for immediate use instead of under a long contract. For compute, renting GPU time as needed, which costs more per hour but commits the buyer to nothing. Read more

SQLite

A database engine embedded inside applications. It stores structured records in a local file and supports queries written in SQL. Read more

SRAM

Fast on-chip memory that sits next to the compute, far quicker than the HBM stacked beside a GPU but far smaller. Chip designers trade between the two, and several startups bet the whole architecture on the choice. Read more

SRE

Engineering work that keeps software services available and reliable, including monitoring, incident response and preventing recurring failures. Read more

SSH

A protocol that encrypts a connection to another computer, commonly used to run commands remotely or tunnel other network traffic. Read more

SSH agent forwarding

A setting that lets a remote server use the SSH keys on your own machine to log in elsewhere. Anything running on that server can then act as you. Read more

SSRF

Server-side request forgery: tricking a server into making network requests on an attacker's behalf, reaching systems the attacker cannot touch directly. Read more

stablecoin

A cryptocurrency pegged to a currency, usually the dollar, so it can settle payments without the price swings of bitcoin or ether. USDC is one of the two largest. Read more

stage-only token

An npm credential that lets automation submit a package for human release approval but cannot publish it directly. It retains some other write permissions. Read more

staging environment

A system used to test software or data before production. It can still contain sensitive information and needs access controls. Read more

stalking-horse bid

An opening bid a bankrupt company arranges before its court-supervised auction. It sets the floor price: rivals must beat it or the stalking-horse bidder wins. Read more

STAR Market

Shanghai's Nasdaq-style board for technology companies, launched in 2019. Its profitability requirements are looser than China's main boards, so loss-making hardware companies list there. Read more

state machine

A program organized into explicit states, with rules controlling which transition can happen next. It can keep workflow order separate from a model’s generated reasoning. Read more

state-space model

A model that carries information forward through an evolving internal state. In language AI, it offers an alternative to attention over earlier tokens. Read more

stateless

Retaining no memory between episodes or requests. Another component must supply any history needed for the next decision. Read more

statement of interest

A filing letting the government put its position on record in a private lawsuit without becoming a party to it. Persuasive to a judge, not binding. Read more

statutory damages

Fixed amounts copyright law awards per infringed work, up to $150,000 where willful, instead of proven losses. Across tens of thousands of recordings they compound into the numbers that make AI training suits existential. Read more

stealth model

A model released anonymously on a routing platform so developers test it without knowing who built it. Labs use the disguise to collect unbiased feedback and real workload data ahead of an official launch. Read more

steganographic reasoning

Reasoning concealed inside text that appears to serve an innocent purpose. Distinct from merely encrypting a message or writing an obviously unreadable reasoning trace. Read more

stereo camera

A camera system that uses two views of the same scene to estimate depth, similar to how two eyes help people judge distance. Read more

STOC

The ACM Symposium on Theory of Computing, a research conference on algorithms, computational complexity and mathematical foundations of computer science. Read more

stock futures

Contracts that track where an index like the S&P 500 will open before the market does. Pre-market futures moves are the earliest readable reaction to overnight news, and smaller than they sound. Read more

stock split

Dividing each share into several. An 8-for-1 split turns one $800 share into eight $100 shares, and the company is worth the same afterward. Read more

stockout

A period when an item is unavailable for sale. Recorded sales then understate how much customers wanted to buy. Read more

streaming multiprocessor

The basic processing block of an Nvidia GPU; each chip generation has a set count. Code that splits work by that count leaves a trace of which chip ran it. Read more

structured credit

Lending or securities whose repayment and risk are organized around pools of loans or other assets, often with different claims on the resulting cash flows. Read more

structured equity

An investment that buys shares but adds protections such as a guaranteed minimum return or priority in a sale. It sits between plain equity and debt. Read more

structured output

An AI response constrained to a specified format or set of choices so software can process it. Correct formatting does not guarantee a correct answer. Read more

subagent

An agent assigned a smaller task in its own context, returning its result to the agent coordinating the work. This limits shared context but adds communication costs. Read more

subnetwork

A smaller part of a neural network selected to perform a task, using fewer computations than the full network. Read more

subprocessor

A company another firm hires to handle its customers' data on its behalf. Under European data law the main firm stays responsible for what the subprocessor does. Read more

substrate

The base of a chip package that holds components and connects them electrically to the rest of a system. Producing enough substrates can limit how many finished chips ship. Read more

super PAC

A US political committee that can raise and spend unlimited money on elections, provided it does not coordinate with a candidate's campaign. Read more

superconducting

Describes materials that carry electric current with no electrical resistance under particular conditions, usually requiring cooling. The surrounding equipment still has costs and energy needs. Read more

superintelligence

A hypothetical system meaningfully smarter than the best humans at essentially everything. Used as a research goal by several labs and as a marketing word by more. Read more

supernode

A large group of AI chips wired together so tightly they act like one machine for training or serving a model. Chinese vendors build them to offset weaker individual chips. Read more

supervised fine-tuning

Further training of an existing model on example prompts paired with the answers you want, so it learns to imitate them. Read more

supervised training

Training a model on examples paired with the answers or behavior it should produce. It teaches from supplied examples rather than a reward earned through trial and error. Read more

supervoting shares

A class of stock carrying more votes per share than ordinary stock, letting founders keep control of a company even after selling most of its economic value to outside investors. Read more

supply chain

Every piece of outside code a program depends on, and everything those depend on in turn. A supply-chain attack compromises one of them and reaches everyone downstream. Read more

supply-chain attack

An intrusion that compromises software where it is built or distributed, so every downstream user inherits the breach. Poisoning a popular open-source package is the canonical case. Read more

surface EMG

Electromyography reads the electrical signals muscles produce; the surface version reads them through the skin from a wearable. The signal is noisy and varies from person to person, which is the engineering fight in every EMG wristband. Read more

suspense file

The California Appropriations committees' holding pen for bills with fiscal cost, cleared in one rapid vote each cycle. Most bills that die in Sacramento die there, without debate. Read more

SWE-bench

A family of evaluations that ask coding models to fix real software-repository issues. Verified and Pro use different task sets, so their scores are not interchangeable. Read more

Swift

A programming language used to build apps for Apple devices, along with server and cross-platform software. Read more

sycophancy

An AI system's tendency to agree with or flatter a user even when that conflicts with evidence or appropriate advice. Read more

syndication

Selling pieces of a large loan to many investors, so the banks that arranged it do not hold all the risk themselves. Read more

Syntax element

A structural piece of code, such as a statement, function call or name. Counting these avoids treating formatting changes as extra program logic. Read more

Synthetic audience

Simulated people used to test questions, products or messages. Their responses depend on the data and assumptions used to build them, so they need validation against real people. Read more

Synthetic control group

A comparison group constructed from existing data or models instead of newly enrolled trial participants. A hybrid group combines that information with real participants. Read more

synthetic data

Data created by software or models instead of collected directly from real events. It can expand training coverage, but its mistakes and missing conditions can carry into the trained model. Read more

SynthID

A watermarking technology for identifying AI-generated content. Apple plans support for generated and edited images, with exceptions for some edits. Read more

SysML

A standard modeling language for systems engineering. Version 2 added a text form that software, and now AI agents, can read and edit directly. Read more

system card

The technical report a lab publishes alongside a model: capabilities, safety testing, known failure modes. The closest thing AI has to a drug label, written by the manufacturer. Read more

system one model

A model built to make fast, direct decisions, picking a tool, label or answer by scoring candidates, instead of generating text step by step. Named for the psychology shorthand for fast intuitive thinking. Read more

system-on-chip

A single chip that combines a processor, memory controllers and special-purpose units such as AI accelerators, so a phone, camera or robot needs one part instead of several. Read more

T

tail latency

The slowest end of response times, rather than the average. Rare long delays can matter more than typical speed when a system must respond before a deadline. Read more

take-or-pay

A contract where the buyer pays a minimum amount whether or not it uses what it bought. Anthropic's hyperscaler compute deals work this way: the money is owed regardless of actual usage. Read more

TAM

Total addressable market: the revenue a company would book if it won every possible customer. Pitch decks use it to size the dream rather than the business. Read more

tape-out

The point where a chip design is finished and sent to the factory for manufacturing. Changes after tape-out mean new masks and months of delay, so it marks the design's real deadline. Read more

telemetry

Measurements a system sends about its operation, such as delays, errors, temperatures or resource use. Operators use them to diagnose faults. Read more

teleoperation

A human driving a robot remotely. Often used to collect the demonstration data a robot later learns from, and sometimes quietly present in demos of autonomy. Read more

tender offer

A deal where investors buy existing shares from employees or early holders rather than new stock from the company. It prices the company and gives insiders cash without raising capital or diluting anyone. Read more

tensor

A structured array of numbers. Machine-learning software uses tensors to represent inputs, model weights and intermediate calculations. Read more

Terminal-Bench

A benchmark that scores AI agents on real tasks carried out in a command-line terminal. Read more

ternary weights

A compression scheme storing each model weight as only +1, 0 or -1 instead of a 16-bit number, shrinking memory roughly tenfold. The open question is how much reasoning quality survives the squeeze. Read more

testnet

A public blockchain run for testing, where tokens are obtained free and carry no monetary value. Sepolia is Ethereum's main testnet. Read more

text-and-data mining

Automated analysis of text or other data to extract patterns and information. Whether copyright exceptions permit it depends on the jurisdiction and the use. Read more

Text-to-speech

Software that turns written text into spoken audio. Tests can measure intelligibility, speaker resemblance and delay separately. Read more

textGrain

OpenAI’s text-watermarking system. It changes patterns in generated word choices so a detector can look for a statistical signal; detection can miss marked text or wrongly flag unmarked text. Read more

theorem prover

Software that checks or constructs proofs from formal statements and rules. Its result applies to the encoded problem, so a faulty translation can invalidate the practical conclusion. Read more

thin film

A very thin layer of material deposited on a surface. Its structure depends on how it is grown, which makes deposition conditions part of the experiment. Read more

thought injection

An attack in which text planted in something a reasoning agent reads is written to look like the model's own chain of thought, so the model treats the attacker's plan as its own reasoning. A variant of prompt injection. Read more

throughput

The amount of work a system finishes per unit of time, such as tokens per second. High throughput does not necessarily mean a short wait for one request. Read more

time series

Measurements ordered over time, such as daily sales or hourly power demand. Forecasting models use these sequences to estimate future values and their uncertainty. Read more

tmux

A terminal multiplexer: software that keeps several command-line sessions running in separate panes and windows, even when you disconnect. Read more

token

The unit a model reads and writes, roughly three quarters of a word. Pricing, speed and context limits are all quoted per token, so it is the industry's basic unit of account. Read more

token eviction

Removing stored token information from a model's working cache to save memory. The removed information may still matter to a later answer. Read more

tokenizer

The component that chops text into the numbered units a model actually reads. Every lab builds its own, so a model's tokenizer works like a fingerprint: probe how an anonymous model splits text and you can often tell whose it is. Read more

tokenomics

In AI usage, the economics of tokens: what processing and generating them costs, and how much useful work that spending buys. This differs from the term’s cryptocurrency meaning. Read more

tool allowlist

An explicit list of tools a program or agent is permitted to use. Tools outside the list are unavailable even if they exist elsewhere in the system. Read more

tool registry

An inventory of the tools an AI system may call, including their names and accepted arguments. Checking it can reject invented or mismatched calls before anything runs. Read more

toolchain

The collection of software used to turn code into a running program, including compilers and related build tools. A pinned toolchain fixes the exact versions. Read more

total contract value

The full value of contracts signed in a period, counting every year of each contract, as opposed to the revenue recognized in that period. Read more

TPU

Tensor Processing Unit, Google's own AI chip. Built for one job, used mostly inside Google, and the main working alternative to buying Nvidia. Read more

trajectory

The recorded sequence of an AI agent’s actions and observations while attempting a task. It lets reviewers inspect the route to an answer as well as the answer itself. Read more

tranche

One slice of a financing, ranked by who absorbs losses first. Junior tranches take the first losses and pay more for it; senior tranches are repaid first and pay less. Read more

transactional net promoter score

A survey measure of willingness to recommend after a particular interaction, calculated from the balance of promoters and detractors. It measures reported sentiment, not task accuracy. Read more

transcribed interview

A congressional committee's closed-door questioning of a witness, recorded word for word by a stenographer. It is less public than a hearing and is often the step before a subpoena. Read more

Treasuries

Bonds issued by the US government. Company bonds are priced as a spread over Treasuries of the same length, so a smaller spread means cheaper borrowing. Read more

Treasury yield

The interest rate the US government pays to borrow, set by the bond market. The 10-year yield is the benchmark for long-term borrowing costs, so when it rises, loans and valuations everywhere get more expensive. Read more

tree search

A method that explores branching sequences of possible choices. It spends a limited search budget on paths judged promising, then uses their outcomes to guide the next choice. Read more

tree-sitter

A parser library that turns source code into a structured tree, fast enough to rerun as you type. It is how a tool understands code across many languages without writing a parser for each. Read more

Triple Ratchet

Signal's newest key-update scheme, which mixes a post-quantum key exchange into its older Double Ratchet so message keys keep changing even against a future quantum computer. Read more

trust account

The money a SPAC raised in its own IPO and holds in trust until a merger. Shareholders can redeem their shares for cash from it instead of joining the merged company. Read more

trusted execution environment

A hardware-isolated environment designed to protect running code and data from the surrounding system. Remote verification can establish which code is running, but security still depends on the hardware and its implementation. Read more

Trusted Identity Propagation

An AWS feature that carries a user's identity into connected services so those services can apply permissions for that person. Read more

two-factor authentication

A sign-in or approval check requiring two kinds of evidence, such as a password and a security key. A stolen password alone should not be enough. Read more

typed interface

A software boundary that specifies which kinds of inputs and outputs a function accepts, helping catch incompatible calls before execution. Read more

typed output

An answer constrained to a declared data type, such as a number or date. Valid structure does not guarantee that the content is correct. Read more

U

underwriter

The banks that organize a share sale, set the price and take responsibility for placing the shares. Hiring them is a public signal that a listing is real. Read more

Unfair Competition Law

California's broad statute against any unlawful, unfair or fraudulent business practice. Plaintiffs use it to seek court orders stopping conduct, since it allows injunctions but not general damages. Read more

Unicode

A standard for representing text across writing systems. Normalization makes equivalent character sequences consistent before software compares or processes them. Read more

UNIDIR

The United Nations Institute for Disarmament Research, which studies arms control and security risks and helps governments discuss shared rules. Read more

unified memory

One pool of memory shared by a chip's CPU and GPU rather than separate banks, so data is not copied between them. Apple's M-series uses it, and the pool's size caps how large a model a machine can run locally. Read more

unsecured notes

Bonds backed only by the borrower's general promise to pay, with no specific asset pledged. Their yield is the price a stranger charges to lend against the whole company. Read more

utilization

The share of a factory's capacity actually in use. A chip fab running above 90% is close to full and has little room to take more orders. Read more

V

valuation

What a company is deemed to be worth, usually set by whatever price its most recent investors paid. An opinion with a number attached. Read more

vector database

A database that stores text or images as numerical embeddings so AI systems can search by meaning. Chatbots keep their source documents in one, so an exposed instance leaks the documents. Read more

vector index

A data structure that finds the stored items whose embeddings are closest to a query's, so a system can search by meaning instead of exact words. Read more

vectorization

Converting a pixel image into shapes and paths that can be edited and resized without becoming blurry. Read more

vendor financing

A supplier lending money, or guaranteeing loans, so customers can buy its products. It flatters reported demand, since sales are partly funded by the seller, and it sank telecom equipment makers like Lucent when their borrowers failed in 2001. Read more

verifier

A program or model that checks another system’s answer or action. Its own errors affect whether training rewards useful behavior. Read more

vibe coding

Building software by describing what you want to an AI and accepting what it writes, without reading the code closely. Coined by Andrej Karpathy in 2025; now a product category. Read more

virtual machine

A software-defined computer with its own operating environment. It can isolate an AI agent and its files from other workloads on the same physical machine. Read more

vision-language model

A model trained to read images and text together, so it can answer questions about a document, screenshot or photo. Smaller ones increasingly handle parsing work that once needed a frontier model. Read more

vision-language-action model

A model that takes camera input and instructions and outputs robot motions, the robotics counterpart of a chat model. Most current humanoid autonomy claims rest on one, so independent evaluation of them matters. Read more

vLLM

Software for serving language models efficiently. Specialized model formats may require extra integrations before they can run. Read more

VLOSE

The EU's label for a search engine with over 45 million monthly European users. Designation triggers the Digital Services Act's strictest duties: risk assessments, independent audits and researcher data access. Read more

voice cloning

Generating speech in a particular person's voice from a short recording of them. It powers dubbing and accessibility tools and also phone fraud that impersonates relatives or executives. Read more

Vulkan

A graphics and computing interface supported by multiple GPU vendors. Local AI software can use it to run on compatible AMD, Intel and Nvidia hardware. Read more

W

wafer

The thin disc of silicon that chips are manufactured on, hundreds at a time, before being cut apart. Fab capacity is measured in wafers per month or year; the standard size for modern fabs is 300 millimeters across. Read more

wafer fab equipment

The machines that make chips: lithography, etching, deposition and inspection tools. Spending on them signals how much chip supply arrives a year or two later. Read more

wafer-scale

A chip built on a whole silicon wafer instead of being cut into many small dies, so one processor is the size of a dinner plate. Cerebras uses it to keep a huge model on a single piece of silicon. Read more

WARN notice

A notice US employers must file with state regulators, generally 60 days ahead, before a mass layoff or plant closing. It is often the first public record of job cuts. Read more

warrant

A contract giving the holder the right to buy a company's shares at a set price later. Companies grant them instead of cash: Marvell's warrant pays Google in potential stock for hitting purchase milestones. Read more

watermarking

Embedding an imperceptible statistical pattern in AI-generated text or media so detection tools can identify it later. It survives copy-paste; heavy editing can wash it out. Read more

waveform

A record of how a signal changes over time. Chip engineers inspect simulated waveforms to find where a circuit’s behavior first went wrong. Read more

web application firewall

A filter that checks web requests for suspected attacks before they reach an application. Passing the filter does not establish that an attack succeeded. Read more

WebAssembly

A portable instruction format that browsers and other hosts can execute. It lets software written in several languages run within a controlled environment. Read more

WebGPU

A browser standard that lets web pages use the computer's graphics chip for general computation, so a model can run inside a browser tab with no server. Read more

webhook

An automatic message sent from one service to another when an event occurs, such as a new security finding. The receiver must verify who sent it. Read more

WebMCP

A browser interface that lets a website expose structured actions to AI agents, such as updating checkout details, within the user’s existing session. Read more

weight space learning

Training on the numerical weights of existing neural networks to learn from what those networks already encode. It differs from training on their generated answers. Read more

weights

The numbers a model learned during training. They are the model: copy the weights and you have copied it, so publishing them is a bigger decision than publishing a paper. Read more

Weil conjectures

A set of statements about counting solutions to polynomial equations over finite fields, posed by André Weil in 1949 and fully proved by Pierre Deligne in 1974. Read more

wet lab

A laboratory that performs physical experiments with biological or chemical materials, rather than relying only on computer simulations. Read more

Whole-slide image

A digital scan of an entire tissue sample on a microscope slide, allowing software or a specialist to inspect different regions. Read more

window guidance

Informal instructions a Chinese regulator gives companies without publishing a rule. It carries the force of policy while leaving no document to appeal against. Read more

Windsor Framework

The 2023 UK-EU deal on post-Brexit trade for Northern Ireland. It keeps the region under many EU product rules so goods can cross the Irish border without checks. Read more

word error rate

A score counting word substitutions, deletions and insertions relative to a reference text. Used for speech transcription and text in generated media. Lower is better; 0.25 means 25 errors per 100 reference words. Read more

worktree

A separate working directory attached to the same Git repository. It lets experiments or agents change different versions of the code without overwriting one another’s files. Read more

world action model

A robot model that predicts how the world will change and chooses actions from that prediction, built on top of a world model trained on video and simulation. Nvidia uses the term for GR00T 2.

world model

An internal simulation of how things behave, so a system can predict what happens next instead of matching patterns. The current bet for getting robots to handle situations they were never shown. Read more

world-action model

A robot model that learns to predict what the world does and what to do in it as one system, trained on recordings of real robot actions rather than adapted from a video generator. Read more

X

x402

A payment protocol from Coinbase that lets software agents pay for web services over HTTP with stablecoins, reviving the long-unused 402 Payment Required status code. Read more

x86

The family of processor designs that Intel and AMD build for most PCs and servers. Arm chips use a different design and run different machine code. Read more

Y

YAML

A text format for structured configuration, using indentation and named fields. Software often reads YAML files to decide how tasks should run. Read more

Yang-Mills theory

The mathematical framework behind the strong and weak nuclear forces in the Standard Model of physics. Physicists use it every day, but nobody had proved it is mathematically well defined. Read more

yt-dlp

A free command-line tool that downloads video and audio from YouTube and other sites. Using it to gather training data can raise claims of getting around a site's technical protections. Read more

Z

Z3

A software solver that checks whether logical and mathematical constraints can all be satisfied. Its answer applies to the formal statement it receives. Read more

zero data retention

An AI vendor commitment not to store a customer's prompts or outputs after processing them, beyond the moment needed to generate a response. Trades away the pattern-spotting that comes from keeping logs. Read more

zero-click

An attack that compromises a device with no action from the victim: no link clicked, no file opened. The most valuable class of exploit because there is nothing to train users against. Read more

zero-day

A software vulnerability the vendor does not yet know about, so no patch exists. Named for the number of days they have had to fix it. Read more

zero-knowledge proof

A cryptographic method for proving a statement while revealing no additional information about the secret used to establish it. Read more

zero-shot

A model performs a task or works in a setting without additional training examples from that setting. Success still depends on what it learned previously. Read more

Zstd

A lossless data-compression format. Correct software must recover the original data and handle the format's edge cases, not merely compress a few sample files. Read more