The only AI glossary you’ll need this year
TL;DR
TechCrunch has updated its AI glossary as a living reference for terms that now show up constantly in product meetings, pitches, and AI coverage. It explains core concepts such as AGI, LLMs, tokens, training, inference, fine-tuning, hallucinations, chain of thought, and reinforcement learning. Newer infrastructure terms like MCP, Mixture of Experts, memory cache, token throughput, coding agents, and RAMageddon show how AI now reaches into hardware, standards, and cost models.
Nauti's Take
Good glossaries are almost infrastructure in 2026. Not because everyone needs to memorize every term, but because vague language hides weak decisions. The useful part here is that TechCrunch goes beyond model vocabulary and includes cost and systems terms like token throughput, memory cache, and RAMageddon.
The blind spot: a glossary sorts meaning, not priority. In practice, the sharper question is which term actually changes how you work, buy, or build.
Briefingshow
AI vocabulary is not just language; it often shapes whether teams build realistic products or chase buzzwords. Understanding the difference between training, inference, tokens, RAG, MCP, and agents makes costs, limits, and risks easier to spot. Precision matters most around terms like AGI and hallucination, where serious strategy and industry theater often sit next to each other.
Sources