TechCrunch AI put together a glossary of the most important AI terms you’ll encounter in 2026, and it’s worth bookmarking. Whether you’re reading headlines about reasoning models or trying to figure out what “distillation” means when it’s not about whiskey, this guide covers the essentials.
📋 Quick Start
What you’ll learn: the plain-English meaning of 10+ core AI terms, from AGI to fine-tuning. No technical background required. By the end, you’ll read AI news without needing to Google every other word.
🤖 AGI (Artificial General Intelligence)
This is the big, fuzzy goal the industry keeps chasing. AGI refers to AI that’s more capable than the average human at many, if not most, tasks.
The tricky part? Nobody agrees on a precise definition. OpenAI CEO Sam Altman recently described it as “the equivalent of a median human that you could hire as a co-worker.” Google DeepMind sees it as “AI that’s at least as capable as humans at most cognitive tasks.”
Why it matters: when companies claim they’re “close to AGI,” understanding the term helps you judge those claims yourself. Even experts at the forefront of AI research are confused by it, according to TechCrunch AI.
🛠️ AI Agent
An AI agent goes beyond a chatbot. It uses AI to perform a series of tasks on your behalf: filing expenses, booking tickets, writing and maintaining code.
The key distinction: agents are autonomous systems that may draw on multiple AI systems to carry out multistep tasks. The infrastructure is still being built out, so “AI agent” means different things to different people right now.
Why it matters: agents are the bridge between AI that answers questions and AI that actually does work for you.
🧠 Chain-of-Thought Reasoning
Think about how you solve math problems. Simple ones you answer instantly. Harder ones need pen and paper with intermediate steps.
Chain-of-thought reasoning works the same way for AI. Instead of jumping straight to an answer, the model breaks a problem into smaller steps. It usually takes longer, but the answer is more likely to be correct, especially for logic or coding tasks.
Reasoning models (like OpenAI’s o-series) are optimized for this approach using reinforcement learning.
⚡ Compute
Compute refers to the computational power that allows AI models to operate. It’s shorthand for the hardware (GPUs, CPUs, TPUs) that forms the bedrock of the modern AI industry.
Why it matters: compute is the fuel. When you hear that training a model cost $100 million, most of that went to compute.
🔬 Deep Learning
A subset of machine learning where algorithms use multi-layered artificial neural networks. This structure lets them make more complex correlations compared to simpler systems like decision trees.
Two things to know:
- Deep learning models identify important characteristics in data themselves, without human engineers defining those features
- They need massive amounts of data (millions of data points or more) and take longer to train, so development costs run higher
The structure draws inspiration from the interconnected pathways of neurons in the human brain.
🎨 Diffusion
The tech behind many art, music, and text-generating AI models. Here’s the clever trick: diffusion systems slowly destroy the structure of data by adding noise until there’s nothing left. Then they learn to reverse the process, restoring the data from noise.
Why it matters: when you generate an image with Midjourney or DALL-E, diffusion is doing the heavy lifting.
📚 Distillation
A technique for extracting knowledge from a large AI model using a “teacher-student” approach:
- Send requests to the large “teacher” model
- Record its outputs (sometimes comparing them against a dataset for accuracy)
- Use those outputs to train a smaller “student” model
The result is a smaller, faster model that approximates the teacher’s behavior. This is likely how OpenAI developed GPT-4 Turbo, TechCrunch AI notes.
One important detail: distilling from a competitor’s model usually violates their terms of service.
🎯 Fine-Tuning
Fine-tuning means further training an AI model to optimize it for a more specific task. You feed in new, specialized data so the model gets better at a particular domain.
Many AI startups take large language models as a starting point and fine-tune them to build commercial products.
What’s Next
This glossary is a living document that TechCrunch AI plans to update regularly as researchers uncover new methods and identify emerging safety risks. Bookmark it, reference it when you hit unfamiliar terms, and use it to cut through the hype.
The original glossary includes additional terms and expanded definitions at TechCrunch AI.