Learn

Understand AI, quickly

Short, honest explainers on the ideas powering the AI ecosystem. No fluff.

Open weights vs open source

Open weights = you can download the model file. Open source = you also get training data, code and a permissive license.

2 min read

Concept

What is RAG?

Retrieval-Augmented Generation lets an LLM look things up before answering, grounding it in your own data.

2 min

Agents

What is MCP?

Model Context Protocol is an open standard for connecting AI assistants to tools, data and services.

3 min

Concept

What is RLHF?

Reinforcement Learning from Human Feedback aligns a model to human preferences after pre-training.

3 min

Agents

How do AI agents work?

Agents are LLMs that plan, call tools, observe results and loop until a goal is reached.

3 min

Concept

What is vector search?

Vector search finds items by meaning, not exact words — by comparing numeric embeddings.

2 min

Models

How does LoRA work?

LoRA fine-tunes a big model cheaply by learning small low-rank update matrices instead of touching all weights.

3 min

Concept

How does diffusion work?

Diffusion models learn to reverse noise: train by destroying an image, then generate by denoising from pure noise.

3 min

Prompting

Prompting 101

Great prompts specify role, task, constraints, format and examples — in that order.

2 min

Terminology

AI terminology cheatsheet

Fast definitions for LLM, token, context window, embedding, fine-tune, agent, MoE, quantization and more.

4 min

Models

What is a Transformer?

The Transformer is the neural-network architecture behind almost every modern LLM — powered by self-attention.

3 min

Models

What is fine-tuning?

Fine-tuning adapts a pre-trained model to your data or task by continuing training on a smaller dataset.

2 min

Terminology

What is a context window?

The context window is the maximum amount of text a model can consider at once — inputs and outputs combined.

2 min

Concept

Why do LLMs hallucinate?

Hallucinations are confident-sounding but false outputs — a natural consequence of models predicting likely text, not truth.

3 min

Models

What is Mixture of Experts (MoE)?

MoE models have many specialist sub-networks but activate only a few per token — big brain, small bill.

3 min

Models

What is quantization?

Quantization shrinks model weights from 16-bit to 8-, 4- or even 2-bit numbers to run faster on smaller hardware.

2 min

Concept

What is multimodal AI?

Multimodal models handle more than text — images, audio, video and even actions — in a single model.

2 min

Terminology

Training vs inference

Training builds the model (very expensive, one time). Inference is running it to answer a request (cheap, but constant).

2 min

Concept

What is AI alignment?

Alignment is the field of making AI systems reliably do what humans actually want — including as they grow more capable.

3 min

Agents

Agentic workflows

Instead of one big prompt, break the task into steps: reflect, plan, use tools, review — often via multiple LLM calls.

3 min

Concept

What is an embedding?

An embedding turns text, images or audio into a list of numbers that capture meaning — the currency of semantic search and RAG.

2 min

Terminology

Open weights vs open source

Open weights = you can download the model file. Open source = you also get training data, code and a permissive license.

2 min

Prompting

Chain-of-thought prompting

Ask the model to 'think step by step' and it will often reason better, especially on math and logic.

2 min

Terminology

What are tokens?

Tokens are the chunks of text LLMs actually see — usually 3-4 characters or half a word each.

2 min