If you’ve ever skimmed an AI article and quietly nodded past terms like LLM, RAG, or RLHF, you’re in good company. TechCrunch AI just published a living glossary built to fix exactly that, and it’s worth walking through. As TechCrunch AI puts it, artificial intelligence is changing the world while inventing a whole new language to describe how it’s doing it. What stands out is that even top researchers disagree on some of these definitions. So this isn’t about you being behind. It’s about a field that hasn’t finished naming itself.
Here’s a practical walkthrough of the core terms, in order, with a note on why each one matters.
Quick Start
You’ll learn nine foundational AI terms and enough context to use them correctly in a conversation. No coding or math background needed. Read it once, and the next AI headline will make far more sense.
The Terms That Trip People Up
AGI (Artificial General Intelligence). A deliberately fuzzy term for AI that’s more capable than the average human across most tasks. Sam Altman once called it the “equivalent of a median human that you could hire as a co-worker.” OpenAI’s charter frames it as “highly autonomous systems that outperform humans at most economically valuable work,” while Google DeepMind says “AI that’s at least as capable as humans at most cognitive tasks.” Why it matters: when companies argue about AGI, they’re often arguing about different definitions.
AI agent. A tool that uses AI to carry out a series of tasks for you, going beyond a basic chatbot. Think filing expenses, booking a table, or writing and maintaining code. The catch, per TechCrunch AI: the term means different things to different people, and the infrastructure to fully deliver on it is still being built.
API endpoints. Picture hidden “buttons” on the back of a piece of software that other programs can press to make it do things. Developers use them to connect apps, like letting one pull data from another. Why this matters now: as AI agents get smarter, they can find and use these endpoints on their own, which opens up powerful and sometimes unexpected automation.
Chain-of-thought reasoning. Some questions a human answers instantly (“which is taller, a giraffe or a cat?”). Others need a pen and paper. If chickens and cows share 40 heads and 120 legs, you write an equation to land on 20 of each. Chain-of-thought means the model breaks a problem into smaller intermediate steps. It’s slower, but the answer is more likely to be right, especially in logic or coding. Reasoning models are built from standard large language models and tuned for this using reinforcement learning.
Agent (the stricter sense). More specific than the loose “AI agent” label: a program that takes actions on its own, step by step, to finish a goal.
Coding agent. A specialized agent for software work. Instead of just suggesting code, it can write, test, and debug autonomously, working across whole codebases and pushing fixes with minimal oversight. TechCrunch AI’s comparison is sharp: it’s like hiring a very fast intern who never sleeps. A human still reviews the work.
Compute. Shorthand for the computational power that lets AI models run. It usually points to hardware like GPUs, CPUs, and TPUs. This is the bedrock the whole industry runs on, which is why chip supply keeps making headlines.
Deep learning. A subset of machine learning built on multi-layered artificial neural networks, loosely inspired by the brain’s neurons. These systems spot important features in data themselves rather than waiting for engineers to define them. The trade-off: they need huge amounts of data (millions of points or more) and take longer and cost more to train.
Diffusion. The tech behind many image, music, and text generators. Inspired by physics, diffusion systems add noise until the original data is destroyed, then learn to reverse that process to create something new. Sugar can’t un-dissolve in coffee, but AI diffusion learns a kind of reverse diffusion to rebuild structure from noise.
Next Steps
Keep the original TechCrunch AI glossary bookmarked, since they update it as the field shifts. Next time you hit an unfamiliar term, look it up before scrolling past. The fastest way to lock these in is to use one in your next conversation about AI. Full definitions and ongoing updates are available at the original source.