[AI GLOSSARY][12 TERMS]

AI Glossary: concepts explained through a working lab

Every entry: a two-sentence definition, how it works, and what our measurements showed. The numbers come from a lab that has been reading and describing AI research since August 2026.

[DATA][RAG]

Vector database

Searches by meaning, not words. Our threshold: 0.764.

[DATA][RAG]

Embedding

Meaning captured in numbers. 18 out of 18 hits.

[SECURITY][LLM]

AI Sycophancy

Why chatbots agree. 25 texts in our database.

[HISTORY][LLM]

Attention Is All You Need

The 2017 paper that gave birth to LLMs. 288 titles in our lab.

[RAG][LLM]

RAG

A chatbot that answers from your documents. 0.08 gr per answer.

[SECURITY][LLM]

Prompt injection

How a chatbot gets hijacked. 548 attacks on our lab chat, 0 leaks.

[SECURITY][LLM]

AI Hallucinations

Why models make things up. Code checks the numbers in our reports.

[AGENTS][LLM][AUTOMATION]

AI Agent

A model that chooses its own steps. 26.3% of arXiv papers in week 43.

[LLM][COSTS]

Tokens and Context Window

How models measure cost and memory. Our question limit: max 500 characters.

[LLM][BASICS]

Large Language Model (LLM)

How the model that writes works. 42.4% of arXiv papers in week 43.

[EVALUATION][LLM]

Model Evaluation

How to test AI with numbers, not guesses. From 0.05 to 0.62 alignment.

[RAG][LLM][COSTS]

Fine-tuning vs RAG

Fine-tune the model or give it documents? Our choice and why.