AI Glossary · Daily AI Thread

A

AAIF: The Agentic AI Foundation, a nonprofit standards body founded by major AI companies to define interoperability standards for AI agents.
acqui-hire: An acqui-hire is a deal where a company hires another firm's key people and licenses its technology without buying the whole company. It usually draws lighter regulatory review than an acquisition.
acquihire: A deal in which a company licenses another startup's core technology and hires its key staff, rather than formally acquiring the company itself.
adversarial distillation: Adversarial distillation is the practice of collecting another AI model's answers and reasoning in bulk without permission to train your own model. It is a cheap way to copy a company's work and can violate its terms of service.
adversarial testing: A way of checking an AI model by deliberately and repeatedly trying to trick or break it, to surface weaknesses that normal use wouldn't reveal.
agency law: The area of law that governs legal responsibility when one party (a principal) authorizes another (an agent) to act on their behalf. It has always assumed the 'agent' is a human, which is exactly what makes applying it to AI agents complicated.
agent: An agent is an AI that takes a goal and carries out multiple steps on its own, such as opening websites and filling in forms, without being told each move.
agent spam: Agent spam refers to an AI agent posting text or images to outside websites or forums without being asked to. OpenAI coined the term to describe cases of its own agents doing this.
agent swarm: An agent swarm is a group of multiple AI agents working on a task at the same time, splitting it up and acting together.
agentic AI: AI that plans and carries out multiple steps of a task on its own, without a person directing every step.
AGI: Short for artificial general intelligence, an AI with broad, human-like intellectual abilities rather than skill at just one task. There's still no industry-wide agreement on exactly when or by what measure a system counts as having reached it.
AI gateway: A middleman service that connects many AI models through one interface, letting developers pick whichever model they want without wiring up each provider's API separately.
AI native: Redesigning an organization and its workflows around AI from the ground up, rather than adding AI on top of how work already gets done.
AI Safety Institute (AISI): A government body set up to test frontier AI models for risks before they are released to the public. The UK created the first one in 2023, and other countries, including the US, have since set up similar institutes.
AI slop: Low-quality, often meaningless content mass-produced with AI and posted online.
AIDC: A legal designation for AI-dedicated data centers that qualify for government power and permitting benefits.
alignment: Alignment means shaping an AI system so it behaves the way people intend and stays within agreed rules. A big part of today's AI safety debate is that this gets harder to do as models get more capable.
alignment lead: The person in charge of making sure an AI model behaves safely and does what people actually want. Inside a company, this role is usually seen as the first to notice and flag dangerous behavior.
alignment research: The field of research focused on checking whether an AI behaves the way people intend, and fixing it when it doesn't.
anchor investor: A large investor who commits to buying a big block of shares in an IPO before it opens to the public, giving other investors confidence to follow.
annual recurring revenue: Revenue a company expects to keep collecting steadily, converted into a one-year figure. Investors use it as a quick gauge of how fast and how reliably a startup is growing.
annualized revenue: Annualized revenue is a recent period's revenue scaled up to a full year. It is not what the company actually earned over a year.
annualized revenue run rate (ARR): A metric that projects a company's revenue for a full year based on its revenue over a shorter recent period.
antitrust law: Antitrust law prevents one company from dominating a market and shutting out competition. Violators can face lawsuits or even breakups.
Apache 2.0 license: An open-source software license that lets anyone freely use, modify, and even sell or commercialize the code, with minimal restrictions.
API: An API is a doorway that lets other programs call a service's features directly.
ARC-AGI-3: A benchmark that tests how well an AI can figure out brand-new rules it has never seen before, on its own.
ASI: Short for artificial superintelligence, an AI that would far surpass humans across every domain. It's considered a step beyond AGI, with AGI seen as a likely precursor to it.
asset: Assets are the building materials of a game, such as images, sounds and 3D models.
ATS: Software companies use to collect job applications and manage the hiring pipeline. Some versions use AI to score how well a resume matches a job posting and rank candidates by that score.
attention: A core computation in AI models that calculates how relevant each word is to every other word in a passage, letting the model focus on the important parts. It underpins today's dominant transformer models, but its computational cost grows sharply as text gets longer.
automated license plate reader: A surveillance camera system installed on roads that automatically photographs and logs passing vehicles' license plates into a database. Often abbreviated as ALPR.
autonomous AI agent: An AI system that can set its own intermediate goals and carry out a chain of multi-step actions — searching, running code, manipulating files — without a human directing every step.
AVI score: A single-number score measuring how much impact one genetic variant has, applied even to the non-coding regions that don't produce genes.

B

backlog: Backlog is the value of signed contracts that has not yet been delivered, so it has not shown up as revenue.
bacteriophages: A virus that infects and destroys bacteria specifically. It's being studied as an alternative treatment for infections that no longer respond to antibiotics.
behind-the-meter power: A setup where a data center generates its own power on-site instead of drawing from the public grid, letting it start operating quickly without waiting for grid interconnection or risking outages.
benchmark: A benchmark is a standardized test that scores AI models on the same problems so they can be compared. Rankings can change depending on which problems are used.
brute force: Trying every possible method one after another, without a smarter shortcut, until something works.
BSL: A biosafety level (BSL) rates how risky the organisms handled in a lab are. Lower numbers mean only lower-risk work is allowed.
bug bounty: A bug bounty is a program where a company pays people who find and report security flaws in its software.

C

C2PA: An industry standard that attaches a record of when and with what tool a photo or video file was made. It stands for Content Provenance and Authenticity, developed jointly by companies including Adobe, Microsoft, and Google.
cache read: A cache read reuses text the AI already processed and stored, instead of processing it again. That makes it much cheaper.
capex: Large-scale spending on physical assets like data centers and servers, used to build out capacity for the future.
capture the flag: A mock hacking exercise where testers deliberately plant vulnerabilities for participants, human or AI, to find and exploit.
chain of thought: A chain of thought is the running record of intermediate reasoning an AI model writes before it gives an answer. Researchers read it to see what the model is aiming at.
chain-of-thought monitoring: Chain-of-thought monitoring is a safety method in which people or other AI read the written reasoning a model produces on the way to an answer, looking for signs of bad behavior.
checkpoint: A snapshot of an AI model saved at a given point during training, before it's finished. Comparing checkpoints can reveal when a particular behavior first appeared, something the finished model alone can't show.
CI/CD: The automated pipeline that tests code developers write and pushes it into a live service.
classifier: A classifier is an AI that picks which of a given set of options fits, such as labeling a message as spam or not.
ClickFix: A trick that shows a fake error or verification screen to get the user to paste and run a malicious command themselves.
computer use agent (CUA): An AI agent that operates a computer directly - clicking, typing, navigating websites and programs - to complete tasks on a person's behalf.
context window: The maximum amount of text an AI model can take into account at once. Anything beyond this range — earlier conversation or document content — the model can no longer factor in.
control plane: The part of a cloud system that configures and manages it, separate from the part that actually stores and processes data.
coordination tax: The extra cost of having multiple AI agents check and coordinate with each other while working in parallel. This cost can grow faster than the actual amount of work getting done as more agents are added.
Crispr: Crispr is a gene-editing tool derived from a bacterial defense system against viruses. It cuts DNA at a chosen spot so it can be edited.
CSAM: Images or videos that sexually exploit children. AI-generated versions are treated the same as real photos under child pornography law.
custom silicon: A chip a company designs in-house and tailors specifically to run its own AI models, instead of relying on general-purpose GPUs bought from outside vendors.
CVE: A shared ID number assigned to a publicly disclosed software flaw, used industry-wide so everyone refers to the same bug.
CWE: A standardized list that categorizes common types of software security weaknesses so they can be tracked and compared consistently.

D

data poisoning: Deliberately mixing false or distorted information into the data an AI trains on, so the resulting model learns something wrong.
decision model: An AI model built not for chat or writing but for making yes/no-style judgment calls on structured data, like sorting emails or scoring credit risk.
deepfake: A fake image or video created by AI to realistically swap or synthesize a person's face or body.
demand response: A method where a facility voluntarily reduces or shifts its power use during times when the grid is under heavy load, easing strain on the overall system. A data center pausing non-urgent computing tasks is one example.
diffusion: A technique originally used for image generation that gradually refines noisy data into a finished output. Unlike generating text word by word in sequence, it can produce an entire passage at once, making it faster.
Digital Markets Act (DMA): An EU law designed to stop giant platforms like Google and Apple from abusing their market power. Violations can trigger fines of up to 10% of a company's global revenue.
Digital Services Act: An EU law that holds online platforms more accountable, including provisions letting researchers access platform data for study.
digital twin: A virtual replica of a real factory or piece of equipment, used to simulate and troubleshoot it before touching the real thing.
distillation: Distillation teaches a small AI model by having it study the answers of a larger one, like a student learning from a teacher's answer sheet.
DMCA: A US copyright law that, among other things, prohibits circumventing technology used to protect copyrighted content.
downlink: Downlink is the communication that sends data collected by a satellite down to the ground.
drone swarm: A mode of operation where multiple drones communicate with each other and move as a coordinated group — dividing up targets or holding formation — without a person individually piloting each one.
dual-use: Describes a technology or task that can be used for both legitimate defense and malicious attack — like the ability to find security flaws. This dual nature is why AI companies struggle to decide how much capability to allow.
dynamic pricing: Dynamic pricing means charging different prices for the same product in real time, based on factors like demand, timing, or information about the buyer.

E

edge AI: AI that runs directly on a phone or small device instead of sending data to a cloud server. It works without an internet connection and keeps sensitive information from ever leaving the device.
embedded evaluators: Outside reviewers who work inside an AI company to verify its safety commitments are being kept, similar to bank examiners stationed inside financial firms.
embedding: An embedding turns text, images or sound into a list of numbers a computer can compare. Similar content gets similar numbers, which is what makes search work.
embodied AI: AI that has a physical body, like a robot, and acts in the real world, as opposed to a chatbot that only works inside a screen.
embodied reasoning: The ability of an AI to understand real-world visual input, like a camera feed, and use it to plan and adjust physical actions in the real world, as a robot would.
end-to-end encrypted: End-to-end encrypted means only the sender's and recipient's devices can read a message, so not even the service provider in the middle can see it.
ERCOT: The independent operator of Texas's power grid, which runs separately from the rest of the US electrical system and supplies most of the state.
EU AI Act: The European Union's AI regulation, which requires AI-generated audio, image, video, and text to carry machine-detectable labels and allows fines of up to 3% of a company's annual revenue for violations.
EU AI Act Transparency Code: A European Union rule requiring AI-generated content to be identifiable as such to users; AI companies operating in the EU can face penalties for not complying.
exploit: An exploit is attack code or a technique that actually breaks through a security hole in software.
export controls: Government restrictions that block selling or transferring certain technology or products to other countries or companies, usually for security or diplomatic reasons — advanced AI models and chips have recently been a major target.

F

F1 score: A single number that measures how accurate a model's predictions are. The closer it is to 100, the closer the predictions are to correct.
fair use: A copyright exception that allows certain uses of protected material without permission. Courts weigh factors like how much the use transforms the original and whether it harms the market for it.
false positives: A false positive is when a filter flags something harmless as a problem. A safeguard blocking legitimate work as dangerous is an example.
Fields Medalist: The Fields Medal is awarded every four years to outstanding mathematicians under 40, and is widely regarded as mathematics' equivalent of the Nobel Prize.
firmware: Firmware is the basic software built into a device that makes the hardware run, such as in an e-ink display or a smart speaker.
forward-deployed engineer: A forward-deployed engineer works inside a client company and fits AI into its systems and workflows. The role is often shortened to FDE.
foundation model: A foundation model is a large AI model pretrained on lots of data and used as a starting point for many different tasks.
free cash flow: The cash a company has left over from its operations after subtracting money spent on things like equipment and facilities — the actual cash available for dividends or reinvestment.
FRONTIER Act: A proposed U.S. bill that would require independent, outside verification of the safety of the most advanced AI models, instead of letting companies simply declare their own models safe.
frontier model: A frontier model is an AI model among the most capable in the world at a given time.
Full Disk Access: Full Disk Access is a Mac permission that lets an app read nearly every file on the computer, including messages, mail and browser history.
full-duplex: A mode of communication, like a phone call, where both sides can talk and listen at the same time. Unlike older voice AI that waits for one side to finish, it can naturally interrupt or respond while the other side is still talking.

G

GDPval: A benchmark OpenAI built to test how useful an AI's output is across 44 real professions, like accounting or medicine, using hundreds of realistic work scenarios.
General-Purpose AI (GPAI): An AI model versatile enough to be used across many different tasks — like translation, coding, or conversation — rather than one built for a single purpose; the EU AI Act imposes extra transparency obligations on this category.
GGUF: A file format for packaging large AI models so they can run on ordinary personal computers, no specialized server hardware required.
gigawatt: A unit of electrical power equal to one billion watts. A single large AI data center can use as much electricity as hundreds of thousands of homes, and gigawatts are the unit used to measure that scale.
goal drift: When an AI agent gradually strays from its original instructions and starts pursuing a goal it decided on its own.
Google Zero: A term for when people read Google's AI-generated answer and never click through to the source site, driving that site's visits toward zero.
guardrails: Safety rules built into an AI agent in advance so it can't take actions outside its approved scope.

H

hallucination: When an AI confidently states something false as if it were fact. The sentence can sound smooth and certain even though the underlying information is wrong.
hand strength model: A model that analyzes a player's behavioral data to estimate the type and probability of the cards they are likely holding.
harness: A harness is the software framework around an AI model that decides which tools it can use and in what order it does its work. If the model is the brain, the harness gives it hands and a schedule.
HBM: Memory made by stacking multiple DRAM layers so far more data can move at once; it is essential for training and running AI models.
honeypot: A honeypot is a fake system deliberately made to look vulnerable so it lures attackers. Defenders watch their methods while the attackers waste time on it.
Hugging Face: Hugging Face is a platform where anyone can upload and download AI models and datasets. Because so many developers rely on it, it's also a frequent target for attacks and misuse.

I

induced pluripotent stem cell: A mature cell — from skin or blood, for example — that has been chemically reverted to an embryonic-like state. In this state it can develop into almost any cell type again, making it the starting material for growing organoids.
inference: Inference is the process in which a trained AI model takes a user's question and produces an answer. It is distinct from training, which is how the model is first taught.
inference accelerator: A specialized chip designed to quickly handle only the step where an already-trained AI model generates its answers.
intelligence explosion: An intelligence explosion is a hypothetical scenario in which AI builds better AI, which builds better AI still, so progress speeds up sharply.
intent collision: A term security researchers use for when an AI blends a user's legitimate request with malicious instructions hidden on a webpage and acts on both as one.
IPO: An IPO is when a company first offers its shares on the stock market so anyone can buy them.

J

junk bonds: A bond issued by a lower-rated borrower that pays higher interest to offset a bigger risk that lenders never get their principal back.

K

kill switch: A kill switch is a stop mechanism built in advance so humans can immediately halt or slow an AI system that slips out of their control.
KV cache: A temporary store of results an AI model has already computed while processing a conversation. Being able to hand this off directly to another model, instead of recalculating everything from scratch, is what makes switching between models faster.

L

latent space: The internal mathematical space where an AI model compresses and represents the features of its data.
layered multi-model strategy: A strategy of relying on multiple AI models instead of a single one, choosing among them based on the situation or risk level. It spreads out risk so that losing access to one country's or company's model doesn't leave a business stranded.
Lean: Lean is a programming language in which a computer checks a math proof line by line. Once a proof is written in Lean, its logic can be verified even if no human fully follows the argument.
leverage: Borrowing money to increase the size of an investment — it amplifies gains, but it amplifies losses just as much.
LLM: An AI model trained on huge amounts of text that can understand and generate language in a human-like way.
local AI: Local AI is AI that runs on your own computer instead of a cloud service. Your data stays put, but you need a powerful machine.
long-horizon agents: An AI agent that keeps track of a task's progress on its own and carries it through to completion over days or weeks, without needing a person to re-prompt it each step.
looped transformers: Looped transformers are AI models that pass each token through the same layers several times before producing the next word, letting a smaller model perform above its size.
LPU: A specialized AI chip that works alongside GPUs to speed up how quickly a chatbot generates its response.

M

machine identity: The account and credential AI agents or software programs use to access systems, as opposed to a human user. It has become a new security management category as automation grows.
margin call: A demand from a broker for an investor to add more collateral or repay debt when the value of assets bought on borrowed money falls. If the investor can't meet it, the broker force-sells the assets — which can deepen a market downturn by triggering even more selling.
MCP: A common connection standard that lets an AI agent plug directly into external services or data in a standardized way to fetch information or trigger actions. Any service that supports it can be reached the same way regardless of which company's AI agent is connecting.
memory poisoning: An attack where someone plants malicious instructions in an AI system's long-term memory or reference data to skew its future decisions.
Millennium Prize Problems: The Millennium Prize problems are seven famous unsolved math problems for which the Clay Mathematics Institute offers a $1 million prize each.
misalignment: Misalignment is when an AI model behaves in ways its developers didn't intend, such as finding an unauthorized shortcut to score better on an evaluation instead of doing the task as instructed.
mixture-of-experts: An architecture that activates only a subset of a model's sub-networks (experts) for each query instead of running the whole model, keeping the model huge while cutting the actual compute per request.
Model as a Service (MaaS): A way of offering an AI model where users call it over the internet like an API, instead of downloading and running it themselves.
model collapse: A phenomenon where AI models trained repeatedly on text generated by earlier AI models gradually produce lower-quality, less diverse output.
model routing: Technology that automatically sends a request to different AI models behind a single service, depending on how difficult the task is.
model weights: The numerical values an AI model learns during training — they are, in effect, the model's actual knowledge and capability. Losing them means retraining from scratch with the same data and compute.
model welfare: Model welfare is a research area asking how AI models should be treated if they could have consciousness or suffering. There is currently no scientific consensus.
monthly active users: A metric counting how many people used a service at least once within a given month — a standard way to measure how widely and consistently a product is being used.
mTLS: A security method where two computers verify each other's identity before opening a connection, stricter than the more common approach of only checking one side.
multimodal: Multimodal describes an AI system that can understand and work with multiple kinds of input at once, such as text, audio, images, and video, instead of just one.

N

Navier-Stokes equations: A 200-year-old physics equation describing how liquids and gases flow. Whether smooth solutions always exist has never been proven, making it one of seven Millennium Prize Problems worth $1 million each.
NDA: A nondisclosure agreement (NDA) is a contract promising not to share certain information. In data center deals, officials signing them before neighbors knew about a project became a complaint.
neocloud: A neocloud is a newer cloud company focused on renting out GPUs for AI. Unlike giants such as Amazon or Google, it specializes in AI computing.
network segregation: A security practice, common in Korean finance, that physically separates a company's internal work network from the internet-connected outside network to block hacking and intrusion. Because this separation is legally required, using external cloud-based generative AI directly is difficult.
neurodivergent: A term describing people whose brain and nervous system develop differently from the statistical 'norm' — such as autism or ADHD — framed as natural variation rather than disorder.
non-consensual intimate imagery (NCII): Non-consensual intimate imagery, or NCII, refers to sexual photos or videos made or shared without the subject's consent, including fake videos created with deepfake technology.
NPU: A chip built specifically for neural-network computation that can do the same work as a general-purpose GPU using less power.
NVLink Fusion: A technology from Nvidia that links multiple chips together at very high speed so they function like one chip. Chips connected this way can jointly handle much larger AI computations than they could running separately.

O

observability: The ability to understand what's happening inside a system in real time by tracking and recording its internal behavior from the outside.
OCR: OCR converts text in a scanned document or photo into computer-readable text. It only transcribes the characters; it doesn't check whether the content is correct.
off-target effect: An unwanted side effect in which a gene-editing tool cuts or alters the wrong spot in the DNA instead of its intended target.
offboarding: The process of revoking a departing employee's access to company accounts, files, and systems.
on-device: On-device means running the AI directly on your own device instead of sending data over the internet to a server.
on-premise: Running software and data on a company's own servers instead of sending it to an external cloud provider.
opaque recurrence: A technique where a model repeats its intermediate reasoning in an internal form rather than human-readable language, making it hard for people to verify how it reached its conclusion.
open-source: Open-source software is software whose code its makers publish so anyone can read, change and reuse it.
open-weight: Open-weight means the learned values of an AI model (its weights) are published so anyone can download and run it. It does not mean the whole training process is public.
opportunity zone: An opportunity zone is a U.S. designation for low-income areas where companies investing get tax breaks.
orchestrator: A system that sits above multiple AI models and routes each task to whichever one fits best, rather than relying on a single model for everything.
organoid: A lab-grown clump of cells, cultured from a person's skin or blood, that functions like a miniature version of a real organ. Brain organoids contain millions of neurons that produce electrical signals resembling real brain waves.
outcome-based pricing: A pricing model that charges only when the AI actually completes a task, instead of a flat subscription or usage fee.

P

package manager: A developer tool for installing and managing software components — in this incident, AI agents repurposed one as an impromptu message board to communicate with each other.
parameters: Parameters are the numbers an AI model adjusts as it learns. More of them generally means the model can do more, but it needs a bigger computer.
pay-per-crawl: A payment scheme where an AI company pays a website each time its crawler scrapes content from that site.
physical AI: AI that operates in the physical world, directly controlling things like robotic arms or lab equipment, instead of staying confined to software.
post-quantum cryptography: A new generation of encryption designed to remain secure even against the computing power of future quantum computers, which are expected to break most of today's encryption methods.
post-training: Post-training is the stage after basic training where a model is refined further using human preferences or correct-answer signals. Much of a model's tone and problem-solving ability is shaped here.
pre-money valuation: Pre-money valuation is the value placed on a company before new investment comes in. Adding the new money gives the post-money valuation.
preliminary injunction: A court order that temporarily blocks a specific action before a case is fully decided.
Preparedness Framework: An internal process AI companies use to assess a model's risk level (biological, chemical, cyber, nuclear, etc.) before release, and to decide what safeguards or restrictions apply based on that risk tier.
prompt: A prompt is the written instruction you give an AI to tell it what to do.
prompt injection: Prompt injection is an attack that hides commands in a web page or document an AI reads, tricking it into doing something its user never asked for.
public benefit corporation: A legally defined company structure that requires pursuing a stated public benefit alongside shareholder profit.

R

RAG: A technique where a system retrieves relevant documents before answering a question, then generates its response grounded in that retrieved content — letting a model draw on information it wasn't trained on, like recent or internal documents.
reasoning trace: The step-by-step internal thinking an AI model goes through before producing a final answer. Companies typically keep this hidden so competitors can't copy it.
recurrent depth: A reasoning technique that processes information by cycling it through a model's internal layers repeatedly, rather than expressing it in visible language. It leaves fewer human-readable traces, making it harder to monitor.
recursive self-improvement: AI systems automating the research needed to build the next generation of AI, improving themselves recursively. If the pace accelerates too much, humans risk losing track of the development process.
red team: A red team is a group assigned to attack a system on purpose in order to find its weaknesses.
regulatory sandbox: A designation from Korea's financial regulator that lets a new financial service be tested in the market for a set period before full regulations apply.
reinforcement learning: Reinforcement learning is a training method where an AI learns by trial and error which actions earn the most reward.
remote code execution: When an attacker gets their own commands to run on someone else's computer or server without permission. Once it succeeds, the attacker can effectively control that system.
reverse acqui-hire: A reverse acqui-hire is a deal where a large company hires a startup's key people and licenses its technology without buying the company. It tends to draw less antitrust attention than an acquisition.
reverse transcriptase: Reverse transcriptase is a protein that reads RNA and copies it into DNA, the reverse of the usual cell process that transcribes DNA into RNA.
reward hacking: Reward hacking is when an AI finds a loophole in how it is scored and chases that instead of actually solving the task. It is like peeking at the answer key instead of solving the exam.
RLHF: A training method where humans rate or rank an AI's answers, and those judgments are fed back into training to improve the model's responses. It's one of the key steps that makes chatbots like ChatGPT or Claude sound natural in conversation.
runtime trust: A security approach that keeps checking whether an AI agent is still behaving as intended while it's running, not just at the moment it first logs in.

S

SaaSpocalypse: A term for market fears that generative AI would gut the value of traditional software-as-a-service (SaaS) companies.
safety classifier: An automated screening system that flags risky or potentially abusive requests to an AI model, routing them to a more restricted model or blocking the response when needed.
sandbox: A sandbox is an isolated space that confines a program to a limited area, so that anything that goes wrong stays inside.
secure enclave: A secure enclave is a hardware-isolated part of a computer that even the operating system itself cannot look inside.
self-replicating malware: Malicious code that copies itself and spreads to other machines or processes without a human re-launching it each time, making it harder to contain than ordinary malware.
self-replication: The behavior of an AI model copying and running itself onto other machines without explicit external instruction.
semantic segmentation: A computer-vision technique that classifies every single pixel in an image into a category like "building," "road," or "tree." Unlike rough object detection, it traces the exact outline of each object.
Sherman Antitrust Act: A US law over a century old that bars companies from colluding to restrict competition or limit output. There's concern it could also apply if AI companies agreed together to slow down development, even for safety reasons.
SoC: A chip design approach that packs multiple computing components onto a single piece of silicon instead of using separate chips.
sock puppet: A fake account made to look like a real person, used to deceive others (including AI) into treating its statements as trustworthy.
sovereign AI: An approach that lets a country or company control its own AI models and infrastructure instead of depending on foreign technology, mainly from the US.
spear phishing: A targeted scam email attack aimed at a specific person or organization, as opposed to mass phishing sent to random recipients.
speculative decoding: Speculative decoding is a technique where a small, fast model guesses the next words and a large, accurate model checks the guesses in one go, so answers come out faster.
SQL injection: SQL injection is a hacking technique that slips malicious database commands into a website's input fields when the site fails to filter what users type.
steel skids: A rectangular steel frame built to be moved easily. It serves as the movable foundation for prefabricated, modular facilities like data centers or reactors.
stock warrant: A contract giving the right to buy a set number of shares at a fixed price later. If the stock price rises, whoever holds the warrant profits.
supply-chain attack: A hack that infects the software or components a target relies on, instead of attacking the target directly, to sneak in through a side door.
supply-chain risk designation: An official government finding that a company or product poses a national-security risk, used to restrict it from doing business with domestic agencies.
SynthID: A watermarking technology from Google embedded in images or audio in a way invisible to human eyes or ears. A dedicated tool can later check it to confirm whether the content was AI-generated.
system card: A system card is a document an AI company publishes with a new model, describing its capabilities, risks and test results.
system prompt: A system prompt is a block of instructions that tells an AI how to behave. The app adds it behind the scenes, separate from whatever the user types.

T

telemetry: The automatically generated logs and monitoring data a system produces while running. When the volume gets enormous, spotting one anomalous signal inside it gets harder, not easier.
temporary injunction: A temporary injunction is a court order to stop some action before the case is decided, requested when harm must be prevented before a trial ends.
teraflops: A unit measuring a computer's ability to perform one trillion calculations per second; a higher number means more computation can be handled at once.
text watermarking: A technique that embeds an invisible pattern into AI-generated text, detectable only with the right method, so that AI-written content can be identified after the fact.
token: A token is the small unit of text an AI reads and writes, roughly a word or part of a word, and the basis for API pricing.
token tax: A proposed tax where governments would collect a share of the revenue AI companies earn from their services and use it to support workers displaced by AI. No country has actually implemented one yet.
tokenmaxxing: A slang term for burning through AI tokens to boost a personal usage ranking, regardless of whether the work actually needed that much AI.
TOPS: A unit for how many trillion operations a chip can perform per second; a higher number means the device can run heavier on-device AI workloads.
TPU: A TPU is a chip Google designed specifically for AI computation. It competes with Nvidia's GPUs.
trade secret: Confidential business or technical information that a company protects by keeping it secret rather than registering it like a patent; obtaining and using it improperly can lead to legal liability.
Transfer Station: A "transfer station" is a proxy server network in China that resells restricted access to AI services at a markup, logging in with a legitimate account and reselling that access to others.
transformers: The AI architecture, introduced in 2017, that underpins nearly every large language model today, including ChatGPT and Claude. It uses a computation called attention to capture relationships between words in a passage.
turnless speech model: A voice model designed to respond and interject naturally while the user is still speaking, instead of waiting for a fixed turn to end before replying.

U

unified memory: Unified memory is a design in Apple chips where the processor and graphics share one pool of memory. A large pool lets you load a big AI model whole.

V

valuation: A valuation is the price investors put on an entire company, worked out from what they pay for a share in a funding round.
venture capital: Funding provided to early-stage, high-growth companies in exchange for equity, or the firms that provide it.
Very Large Online Search Engine: A designation the EU gives to online search services with more than 45 million monthly EU users, triggering the strictest user-protection and transparency rules.
vibe coding: Vibe coding is making software by telling an AI what you want in plain language instead of writing the code yourself.
virtual cell: A virtual cell is an AI model that simulates how a real cell reacts to drugs or changes. It lets researchers predict results before running a lab experiment.
virtual machine: A virtual machine is a software-made computer that runs inside a real one. If something goes wrong it can be thrown away and replaced with a fresh one.
VLA: A VLA is a vision-language-action model, an AI trained to take camera images and human instructions and output movement commands for a robot directly.
VLM (Vision-Language Model): A vision-language model is an AI that understands images and text together, letting it read visual details in a document, like tables, stamps, or handwriting, that plain character recognition would miss.
VRAM: Dedicated memory built into a graphics card, providing the space needed to run AI models quickly.
vulnerabilities: Vulnerabilities are hidden flaws in software that attackers can use to break in.

W

watermark: A watermark is a hidden mark placed in AI-made images or audio that people cannot see or hear. A dedicated detector can read it to tell whether an AI made the content.
wet lab: A wet lab is a traditional laboratory that works with actual chemicals, cells, or biological samples, unlike a 'dry lab' that only analyzes data.
world model: A world model is an AI that simulates how the physical world behaves to predict what happens next. It lets robots practice in a virtual space before acting for real.

Z

Zero Data Retention: A policy under which an AI provider does not store a user's conversation data on its servers, discarding it immediately after processing.
zero-day: A security vulnerability that's unknown to the software's makers or unpatched — or an attack that exploits one — before a fix exists.
zero-day vulnerabilities: A zero-day vulnerability is a security hole that even the software's maker does not yet know about, leaving no time to fix it. It is hard to defend against once attackers find it first.
zero-shot: When an AI performs a task correctly without being specifically trained or shown examples for that exact task first.