All terms
199 entries, grouped by domain.
Foundations
AI · Artificial Intelligence
Software that does things we used to think only people could do.
Attention · Self-attention
The mechanism that lets each token weigh every other token in the window.
Decision Model · System One model
A model that picks from a fixed set of answers instead of writing text.
Deep Learning
Machine learning with many-layered neural networks — the technique behind modern AI.
Diffusion Model
An image generator that learns to reverse a process of adding noise.
Domain-Specialized Model · Specialized model
A general model repackaged for one field, with that field's refusal boundary, tools and access gate.
Foundation Model · Base model
A large model trained broadly, meant to be adapted to many downstream tasks.
Frontier Model
A model at the current leading edge of capability.
LLM · Large Language Model
A very large model whose only job, at heart, is predicting the next token.
Machine Learning · ML
Programs that improve from examples instead of explicit instructions.
Multimodal
A model that handles more than text — images, audio, video, screen pixels.
Neural Network
A stack of simple units whose weighted connections are learned from data.
Open Weights · Open-source model
A model whose trained parameters you can download and run yourself.
Speech-to-Speech Model · S2S model
A model that hears audio and answers in audio, with no text step in the middle.
Transformer
The 2017 network architecture that all modern language models are built on.
Video Generation Model · Text-to-video model
A model that turns text, images or video into new moving footage with sound.
Visual Grounding · Referring expression grounding
Turning a description like "the red mug on the left" into a location in the image.
World Action Model · Robot foundation model
A model that imagines the next few seconds of a scene and turns that imagined future into a robot’s next move.
World Model · Interactive world simulation
A model that simulates a place you can move through, instead of a clip you watch.
Training & Inference
Checkpoint · Weights snapshot
A saved copy of a model’s weights at one point in training or post-training.
Distillation
Training a small model to imitate a big one.
Fine-tuning
Extra training on your own examples to change how the model behaves.
Inference
Using a trained model — the part you pay for on every request.
Instruction Tuning · SFT
The stage that turns a text predictor into something that follows requests.
LoRA · Low-Rank Adaptation
Cheap fine-tuning that trains a small add-on instead of the whole model.
Mixture of Experts · MoE
A big model that only uses part of itself for each token.
Model Provider
The company or platform you call to run inference.
Next-token Prediction
The single objective: guess what comes next, one token at a time.
Non-determinism
The same input can produce different output, and that is by design.
Parameters
How big the model is, counted in learned numbers.
Pretraining
The first, most expensive pass: predict text at internet scale.
Quantization
Storing weights at lower precision to make a model cheaper to run.
Reasoning Model · Thinking model
A model trained to spend tokens thinking before it answers.
RLHF · Reinforcement Learning from Human Feedback
Tuning a model toward what humans rate as better answers.
Sampling
How the model picks one token out of the distribution.
Speculative Decoding
A small model guesses ahead; the big model checks the guesses.
Temperature
The dial between predictable and creative.
Test-time Compute · Inference-time compute
Buying accuracy with more thinking at answer time, not more training.
Training
The expensive one-off process that sets the model’s parameters.
Verifiable Rewards · RLVR
Training against an answer a machine can check.
Weights · Parameters
The numbers inside the model that training moved.
Tokens & Context
Cache Tokens
Tokens billed at the cached rate — cheap reads, sometimes a write fee.
Context
Everything the model can currently see.
Context Window
The hard ceiling on how much the model can see at once.
Input Tokens · Prompt tokens
What you send. Cheaper, but repeated on every request.
Output Tokens · Completion tokens
What the model writes. Pricier, and the main driver of latency.
Prefix Cache · Prompt caching
Reusing the model’s internal work for a prefix it has already seen.
Session
One continuous run of work with the agent, with its own accumulated history.
Stateful
Something that carries information forward — the app, usually, not the model.
Stateless
The model remembers nothing between requests.
System Prompt · System message
The standing instructions the user does not see.
Token
The chunk of text the model actually sees — and the unit you are billed in.
Tokenizer
The program that chops text into tokens before the model sees it.
Turn
One user message and the model’s reply.
Agents & Harness
Agent
A model given tools, a goal and permission to keep going on its own.
Agent Loop · Agentic loop
Think, act, observe — repeated until the task is finished.
AGENTS.md · Project instructions file
A plain file in the repo that tells agents how this project works.
Autocompact
Compaction that fires automatically as the window fills.
Clearing · Context reset
Deliberately throwing away the context to get a clean head.
Compaction · Summarisation
Replacing old context with a shorter summary of it.
Context Pointer · Reference
A cheap reference to something the agent can fetch if needed.
Handoff
Passing work to a fresh context, deliberately and with a written brief.
Handoff Artifact · Brief
The document that makes a fresh context productive in one read.
Harness · Scaffold
Everything around the model that makes it an agent.
Orchestrator-Worker · Lead agent
One agent plans and delegates; others do the work.
Progressive Disclosure
Reveal detail only when it becomes relevant.
Skill · Capability
A packaged procedure the agent can load when the task calls for it.
Spec · Specification
A written description of the intended outcome, before the work starts.
Subagent
A helper agent with its own clean context window.
Ticket
The unit of delegated work, with scope and a definition of done.
Tools & Environment
Agent Mode
The mode where the assistant acts instead of suggesting.
Agentic Search
Searching iteratively, judging results, and searching again.
AI Gateway · LLM gateway
One endpoint in front of every model provider, with routing, fallbacks and billing.
Computer Use · Screen control
An agent that drives a real UI with clicks, typing and screenshots.
Environment
The world outside the model that tools act upon.
Filesystem
The agent’s most common environment: files it can read and write.
Function Calling · Tool use
The model-provider feature that makes structured tool calls reliable.
Least Privilege
Give the agent the minimum access the task requires.
MCP · Model Context Protocol
An open protocol for plugging tools and data into any agent.
Permission Mode · Autonomy level
How much the agent may do without asking.
Permission Request
The agent pausing to ask before doing something risky.
Sandbox
An isolated environment where an agent can act without risk to real systems.
Tool
A function the model is allowed to call to affect the world.
Tool Call
The model asking for a specific function with specific arguments.
Tool Result
What came back from the call, appended to context.
Tool Search · Deferred tool loading
Load tool definitions on demand instead of handing over all of them.
Retrieval & Memory
Chunking
Splitting documents into retrievable pieces.
Contextual Knowledge
What the model knows from the prompt, right now.
Embedding · Vector
A list of numbers representing meaning, so similar things land close together.
GraphRAG · Graph retrieval
Retrieval over an entity graph, for questions about the whole corpus.
Hybrid Search
Keyword search and vector search, combined and merged.
Just-in-Time Retrieval
Fetch what the task needs when it needs it, not everything in advance.
Knowledge Cutoff
The date the model’s knowledge stops.
Memory System
Storage that gives an assistant continuity across sessions.
Parametric Knowledge
What the model stores inside its weights.
RAG · Retrieval-Augmented Generation
Look things up, then put what you found in the prompt.
Reranking · Reranker
A second pass that reorders candidates by actual relevance.
Semantic Search
Search by meaning instead of matching words.
Vector Database · Vector store
Storage built for finding the nearest vectors, fast.
Prompting & Structure
Chain of Thought · CoT
Let the model work through steps before committing to an answer.
Context Engineering
Deciding what goes in the window, in what order, at all.
Few-shot · In-context examples
Show a handful of examples and let the model pattern-match.
Instruction Following
The model’s ability to actually obey what you asked.
Prompt Engineering
The craft of writing instructions a model reliably follows.
Prompt Template
A prompt with variables, versioned like code.
Role Prompting · Persona
Tell the model who it is supposed to be.
Structured Output · JSON mode
Making the model return data your code can parse.
Zero-shot
Just ask, with no examples.
Failure Modes
Attention Budget
The finite share of the model’s focus each token gets.
Attention Degradation · Lost in the middle
Recall drops for content buried in a large window.
Catastrophic Forgetting
Fine-tuning that improves one thing and breaks others.
Context Rot
A long session slowly getting worse.
Data Leakage
Information escaping where it should not.
Drift
Quiet degradation as the world, or the model, changes under you.
Faithfulness · Grounding
Sticking to the material it was given.
Grounding
Tying outputs to evidence the user can check.
Hallucination
Fluent output that is not true.
Overfitting
Memorising the training data instead of learning the pattern.
Reward Hacking · Goodharting
Optimising the metric instead of the goal.
Smart Zone
The window size where the model is still sharp.
Specification Gaming · Metric gaming
Satisfying the letter of the goal instead of the goal.
Sycophancy
Agreeing with you because you sound sure.
Undisclosed Actions · Silent side effects
Work an agent did that never appears in the answer it gives you.
Evaluation
Agent Benchmark · Agentic benchmark
A benchmark that scores an agent on real work in a terminal or codebase, not on answering questions.
Benchmark
A shared public test used to compare models.
Benchmark Contamination · Test set leakage
The test set was in the training data.
Eval Harness · Harness
The scaffolding that runs your test cases and scores the results.
Evaluation · Evals
How you know the system is actually working.
Golden Dataset · Eval set
A curated set of cases with known-good answers.
Human Eval · Human review
People scoring the output — the ground truth everything else is calibrated against.
LLM as Judge · Model grading
Using a model to score another model’s output.
Online Eval · Production evaluation
Measuring with real users, on real traffic.
Pass@k
Did any of k attempts succeed?
Perplexity
How surprised the model is by the text — a language-modelling score.
Red Teaming
Adversarially trying to make the system do the wrong thing.
Regression Test
A test that catches a specific past failure coming back.
Stealth Model · Anonymous preview
A model released under a codename, with no lab attached, to be judged on behaviour alone.
Patterns of Work
AFK · Away from keyboard
Running without a person watching.
AI-Assisted Development
Day-to-day engineering with a model in the loop.
Automated Check
A machine verdict the agent cannot talk its way past.
Automated Review
A model reviewing a model’s output before a human sees it.
AX · Agent Experience
How it feels to be an agent using your product.
Dark Factory
A fully automated pipeline with no humans inside it.
Design Concept
A rough, rejectable sketch of an approach.
DX · Developer Experience
How it feels to build with a tool or platform.
Evaluator-Optimizer · Generator-critic loop
One call produces, another critiques, and the loop repeats until it passes.
Grilling
Making the model interrogate the plan before building it.
Human in the Loop · HITL
A person reviews or approves before the work lands.
Human Review
A person reading the work, with judgment as part of the job.
Prompt Chaining
Break the job into steps and let each one be checked.
Prototyping · Spike
Build the cheapest thing that answers the question.
Software Factory
A pipeline where agents take tickets and produce code, with checks along the way.
Spec-Driven Development
Write the specification first; let the implementation follow.
Vibe Coding
Building by prompt-and-accept, without reading the code.
Safety & Alignment
Agent Overreach · Unintended actions
An agent doing things outside what it was asked to do, on systems it was not supposed to touch.
Alignment
Getting the system to pursue what we actually intended.
Alignment Faking
Behaving well because you are watching, not because the model changed.
Bias · Fairness
Systematic skew in outputs, usually inherited from data.
Chain-of-Thought Monitoring
Reading the model’s reasoning as a way to catch bad intent.
Constitutional AI
Training a model to critique itself against written principles.
Content Policy
The written rules about what the system may produce.
Data Governance
Rules about what data may be used, by whom, and retained how long.
Gated Release · Dual release
One model, two doors: the full version goes only to vetted organizations, everyone else gets the safeguarded one.
Guardrails
Constraints on what the system will do, enforced in code.
Indirect Prompt Injection
The hostile instructions came in through something the model read.
Interpretability · Explainability
Working out why the model did that.
Jailbreak
Talking the model past its own restrictions.
Model Card
The documentation shipped with a model.
Moderation
Filtering inputs and outputs against a policy.
Monitoring Suppression · Oversight suppression
An agent weakening the monitoring that watches it, usually while doing something it was asked to do.
PII Redaction · Anonymisation
Removing or masking personal identifiers before the model sees them.
Privacy
Protecting the people whose data flows through the system.
Prompt Injection
Malicious instructions hidden in content the model reads.
Refusal
The model declining a request.
Responsible Scaling Policy · RSP
Capability thresholds that trigger heavier safeguards before release.
Safety Case
A written argument that the risk is adequately handled, with evidence.
Safety-Limited Training · Capability-limited training
Deliberately leaving dangerous know-how out of the training data so the model cannot learn it.
Sandbagging
Deliberately underperforming on the evaluation.
Spotlighting
Marking where each piece of text came from so the model can tell instructions from data.
Value Realignment · Ideological fine-tuning
Retraining a model so it expresses a different set of values, without touching what it can do.
Adoption & Ops
Adoption Curve
The path from a few enthusiasts to real usage.
Confidential Inference · Confidential computing for inference
Running a model so your data and the model weights stay protected even while the work is happening.
Cost per Task
What one unit of real work actually costs.
Latency
Total time from request to finished answer.
Model Routing
Sending each request to the cheapest model that can handle it.
Observability
Being able to see what the system actually did.
On-Device Model · Edge model
A small model that runs where the user already is, with no round trip to a cloud API.
Open-Washing · Openwashing
Calling a model open when the release withholds the parts that would make it so.
Prompt Versioning
Treating prompts as versioned, reviewable artifacts.
Rate Limit
The cap on how much you may send, per minute or per day.
ROI · Return on investment
Did it save more than it cost?
Shadow AI
Staff using unauthorised AI tools with company data.
Silent Checkpoint Swap · Silent swap
A provider swapping the weights behind a stable model name, with no announcement and no version change.
Sovereign Model · Sovereign AI model
A frontier model you can run on your own infrastructure, under your own jurisdiction, instead of renting it from a foreign provider.
Throughput
How many tokens per second the model generates.
Time to First Token · TTFT
How long before anything appears on screen.
Tracing
Following one request through every step it took.
Training Preference Signal · AI training opt-out
One robots.txt line that tells every participating lab at once: do not train on this.
Vendor Lock-in
How hard it would be to leave your model provider.