How AI actually works, and the models behind it
The foundation models behind today's AI tools, and the concepts behind how they work — tracked, benchmarked, and explained in plain language.
Models
xAI
Grok (4.5)
xAI's model family, notable for tight integration with X (formerly Twitter) and one of the largest disclosed training clusters in the industry.
Anthropic
Claude (Sonnet 5)
Anthropic's model family, built with a heavy emphasis on Constitutional AI — currently the Sonnet 5, Opus 4.8, Fable 5, and Haiku 4.5 lineup.
DeepSeek
DeepSeek (V4)
The Chinese open-weight model family that shook up the industry by matching closed frontier models at a fraction of the reported training cost.
OpenAI
GPT (GPT-5.5)
OpenAI's flagship model family — the engine behind ChatGPT's default experience.
Meta Superintelligence Labs
Muse Spark
Meta's first proprietary, closed-weight flagship model — a striking departure from the open-weight Llama line.
Mistral AI
Mistral (Medium 3.5)
The French AI lab's model lineup — closed-weight flagships alongside some of the most-used open models in the industry.
Google DeepMind
Gemini (3.1 Pro)
Google DeepMind's natively multimodal model family, notable for very long context windows and strong reasoning benchmarks.
Alibaba
Qwen (3.5)
Alibaba's open-weight model family, shipped in an unusually wide range of sizes from mobile-friendly to frontier-scale.
Meta
Llama (4 Scout / Maverick)
Meta's open-weight model family — widely used as the base for countless fine-tuned and self-hosted AI products.
Foundations
How Transformers Work
The neural network architecture behind every modern LLM — attention, tokens, and why it scales so well.
What Are Tokens?
The chunks of text an LLM actually reads and writes — and why they don't map 1:1 to words.
What Is a Context Window?
The model's working memory — how much text it can actually see at once, and what happens when you exceed it.
Training
What Is RLHF?
How raw next-token predictors get turned into helpful, well-behaved assistants.
Pretraining: How Models Learn Language
The first, most expensive training stage — where a model learns to predict text from a huge unlabeled corpus.
Fine-Tuning vs Prompting
Two very different ways to get a general-purpose model to behave the way you want for your specific use case.