AI glossary: the vocabulary
of 2023–2026.
Not a dictionary — a map of how the field's language actually evolved. 400 terms that shaped the AI conversation: which ones emerged when, who coined them, which became standards and which stayed hype, and how they connect.
Built by cross-referencing 22 frontier AI models across three waves, then reviewed against 1230 external references. Entries show their available sources and maturity basis; coverage varies by term.
See how the ideas connect.
Explore one question at a time. Select a term to follow its connections, or take a map into your next presentation.
What information does the model need for the next step?
Context engineering brings instructions, retrieved evidence, history and working state together at inference time.
A larger context window is capacity, not a promise of reliable recall. Prompt caching reuses computation; it does not remove context.
Read the connections (6)
- Prompt engineering→ shapes instructions for →Context engineering
- RAG→ supplies evidence to →Context engineering
- Long-context models→ provides capacity for →Context engineering
- Context compaction→ compresses history for →Context engineering
- Active context curation→ selects information for →Context engineering
- Prompt caching→ reuses prefix compute in →Context engineering
- Prompt engineering→ shapes instructions for →Context engineering
- RAG→ supplies evidence to →Context engineering
- Long-context models→ provides capacity for →Context engineering
- Context compaction→ compresses history for →Context engineering
- Active context curation→ selects information for →Context engineering
- Prompt caching→ reuses prefix compute in →Context engineering
How the vocabulary grew
Each wave was generated blind to add only new terms — so they show genuine expansion, not repetition.
Signal vs hype
A term's maturity is how settled it is — not how important. Click a band to filter.
The dictionary
Definitions are AI-generated paraphrases (not direct quotes) reconciled across 22 models and verified against sources; ⚠ marks emerging terms still settling into the discourse. Content CC BY 4.0.