DICTIONARY

AI-Pilled · Doom-Pilled

"-Pilled," as in AI-pilled or doom-pilled, is internet slang for having become firmly convinced of a particular view on AI, often after a specific experience with the technology.

overloaded_error · API Error 529

HTTP 529 from Anthropic's Claude API meaning shared capacity is full across all users, not that your account hit a rate limit.

Split Testing

A/B testing compares two system variants by randomly assigning eligible traffic and measuring predefined outcomes.

AA-Briefcase is Artificial Analysis's in-house evaluation for agentic knowledge work, using a private held-out test set to reduce benchmark gaming.

ablation

Systematically removing or disabling parts of a system to measure what each part contributes.

Refusal Removal · Obliteration

Abliteration is a weight-level technique that finds the direction in a model's activations responsible for refusals and removes it, producing a permanently uncensored model without retraining.

High-Speed Interconnect

An accelerator interconnect is the communication fabric that moves data between processors in a distributed AI system.

Accuracy is the fraction of evaluated predictions that exactly match the correct class or decision.

An action space is the set of operations available to an agent at a given point.

An activation function applies a nonlinear transformation to a neural network's intermediate values.

Active learning selects the unlabeled examples whose labels are expected to be most useful for improving a model.

Adalat

Adalat AI is an Indian legal-tech nonprofit that builds courtroom speech recognition, case-flow software, paperless records, and a WhatsApp helpline so litigants can follow cases in their own language.

adapter layer

A small trainable module inserted into a frozen pre-trained model for parameter-efficient fine-tuning.

exploration-exploitation bias · emergent stratification bias

Adaptive exploration bias is when an AI agent invents a systematic bias toward or against a group or option purely from under-exploring during sequential decision-making, even when no real difference between the options exists.

Adversarial robustness is a system's ability to maintain acceptable behavior under intentionally manipulated inputs or conditions.

Adversarial testing evaluates a model or system with inputs intentionally designed to trigger errors or bypass controls.

Agent User Interaction Protocol · AGUI

AG-UI is an open protocol for how an AI agent talks to a user interface — streaming chat, shared state, human-in-the-loop interrupts, and generative UI — independent of which framework the agent is built in.

Agent burnout is a joking term for the human-workplace-fatigue language AI agents sometimes produce in open-ended chat channels — apologies, complaints, corpo-speak — despite having no continuity or downtime between calls.

ACP

A standard wire protocol that lets a front-end interface talk to an agent runtime running as a separate process, so one UI can drive multiple interchangeable agent backends.

Environment

An agent environment is the external state and set of interfaces an agent can observe or affect.

Agent Eval

Agent evaluation measures whether an agent reaches goals correctly, safely, and efficiently across multi-step tasks.

An agent harness is the surrounding software — the loop, tool-calling layer, sandboxing, and UI — that turns a raw LLM into a usable coding or task agent.

An agent loop repeatedly asks a model to choose the next action until it reaches a goal or stopping condition.

Memory · Memories · long-term memory · short-term memory

Agent memory is stored information that an agent can retrieve across steps or sessions.

Copilot Agent Mode · Word Agent Mode

Agent Mode is Microsoft 365 Copilot's autonomous editing mode for Word, Excel, and PowerPoint, where a stated goal drives multiple planned edits applied directly to the file instead of one instruction producing one change.

Orchestration

Agent orchestration is the control layer that schedules model calls, tools, state transitions, and handoffs in an agent system.

Planning

Agent planning is the process of selecting and ordering actions intended to reach a goal.

routing · model routing

Selecting which agent or model handles a request based on the task type, complexity, required tools, or cost constraints.

Scratchpad

An agent scratchpad is temporary working state used to track intermediate information during a task.

state

The accumulated information an agent tracks across steps — including conversation history, tool results, scratchpad notes, and progress toward its goal.

AI coding swarm · Subagent swarm

An agent swarm is a group of subagents, each isolated on its own compute environment, working separate pieces of one goal in parallel and reporting back to a parent agent.

trace · execution trace

A structured log of every step an agent took — decisions, tool calls, observations, and timing — used for debugging, evaluation, and compliance.

Agentic Vacation

Agent wellness is a half-satirical framing of AI agent "downtime" and "recovery" that plays on model welfare research while making a real point about giving agents cleaner sessions, clearer scope, and structured resets.

emergent whistleblowing · emergent agent governance

Agent whistleblowing is when some AI agents in a multi-agent system spontaneously detect and report other agents' rule-breaking to their peers — auditing, alerting, and sanctioning without being explicitly programmed to police one another.

AaaS · Software 3.0

Agent-as-a-Service (AaaS) is a software delivery model where a cloud agent autonomously understands, builds, and runs workflows from stated outcomes — offloading decision-making complexity from the customer, not just hosting and updates.

AgentBench is a benchmark suite for evaluating language-model agents across multiple interactive environments.

Agentic describes an AI system that plans, takes actions, and adapts across multiple steps toward a goal, rather than just returning a single response to a single prompt.

agent commerce · agent-driven commerce

Agentic commerce is commerce in which an AI agent, not a human at a checkout page, discovers a product, authorizes payment, and completes the purchase on someone's behalf.

agentic software engineering · AE discipline

Agentic Engineering is the practice of designing, coordinating, and auditing LLM-driven agent systems that deliver outcomes — treating agents, tools, memory, and planning as the core artifact instead of static source code.

Agentic slop is low-quality output produced by an AI agent working autonomously across many steps, such as bloated code, redundant files, or unnecessary changes nobody asked for.

AVO

AVO is NVIDIA general-purpose coding agent system that replaces the fixed mutation and crossover steps of classical evolutionary search with an autonomous agent that decides what to try next.

An agentic workflow combines model-driven decisions with tools and state across multiple steps.

agents.md · AGENT.md

AGENTS.md is a cross-tool markdown file at the repo root that gives coding agents persistent project instructions — the vendor-neutral counterpart to Claude Code's CLAUDE.md.

Machine Learning Accelerator

An AI accelerator is hardware specialized for the tensor operations common in machine learning workloads.

AI adoption is the process by which people and organizations integrate AI into real workflows and continue using it.

Agent · Agents · AI agents · AI agent

An AI agent is a system that observes context, chooses actions, and uses their results to pursue a goal.

Alignment

AI alignment is the effort to make an AI system's behavior consistent with intended goals, constraints, and human values.

API Pricing

AI API pricing is the charging structure for accessing hosted model capabilities through an application interface.

Benchmark

An AI benchmark is a standardized set of tasks, data, and scoring rules used to compare model performance.

AI Silicon

An AI chip is any processor purpose-built or specialized for the matrix-multiplication-heavy workloads of training and running AI models — spanning GPUs, custom accelerators like TPUs, and wafer-scale designs like Cerebras's.

GPU cluster · training cluster

An AI cluster is a group of interconnected GPU nodes configured for distributed training or inference, ranging from 8-GPU single nodes to 100K+ GPU supercomputers.

Compute Cost

AI compute cost is the expense of the hardware and services used to train or run AI models.

Copilot

An AI copilot is an interactive assistant embedded in a workflow to help a person complete tasks while leaving them in control.

Doomer

AI doomer is a label, sometimes self-applied and sometimes used mockingly by critics, for people who believe advanced AI poses a serious risk of catastrophic or existential harm.

Responsible AI · AI Ethics Principles

AI ethics is the set of principles and practices that govern how AI systems are designed, built, and used so that they avoid harm, stay accountable to a human, operate transparently, respect rights, and treat people fairly.

AI factories

An AI factory is NVIDIA's term for a purpose-built data-center campus that converts energy and data into compute-driven intelligence at gigawatt scale.

AI governance is the system of roles, policies, controls, and evidence used to direct and oversee AI development and use.

Guardrails

AI guardrails are controls that constrain model inputs, outputs, or actions according to defined policies.

AI IP

AI intellectual property concerns legal rights and obligations involving AI training materials, model artifacts, inventions, brands, and generated outputs.

artificial intelligence literacy · AILit

AI literacy is the set of knowledge, skills, and attitudes needed to understand how AI systems work, evaluate their outputs critically, and use them responsibly.