How LLMs Work: Interactive visual guide to understanding large language models

What it does

How LLMs Work is an interactive visual guide explaining how large language models are built and trained. The site walks through the complete pipeline from raw internet text to a conversational AI assistant, covering pre-training stages like data collection, filtering, tokenization, and model training. It includes live demonstrations of LLM responses and breakdowns of representative scale figures from frontier models (15 trillion tokens, 405 billion parameters, 44 terabytes of training data, 100K token vocabulary).

Who it is for

The guide targets people who want to understand the technical foundations of large language models without specialized machine learning background. It works for curious technologists, founders evaluating LLM-based business ideas, students of AI, and anyone trying to grasp how systems like ChatGPT actually function. The content is accessible rather than research-focused.

Pricing

The site does not show prices. The content appears to be freely available.

How it stands out

The guide is built as a single HTML file generated from Andrej Karpathy's "Intro to Large Language Models" lecture using Claude Code. This approach reduces friction—no loading times between sections, no external dependencies, and the entire experience is self-contained. The visual interactivity distinguishes it from reading a transcript or watching a video; users can click through stages of the training pipeline, animate processes, and see live LLM responses embedded in the explanations. The focus on contemporary figures (circa 2024) grounds abstract concepts in real model scales.

What a founder should check

A founder considering a similar explainer product should verify: (1) whether demand exists for interactive educational content on AI topics, or if the market prefers video tutorials, text blogs, or formal courses that already serve this audience; (2) how much stickiness or repeat engagement these guides generate—the creator notes revisiting this content themselves, but quantifying user return rates matters for building a sustainable business; (3) whether there is a defensible moat beyond better visualization, such as proprietary data, instructor reputation, certification, or integration with developer tools, since educational content on public topics faces commoditization pressure.

Thinking of building something like this?

Every launch here is a competitor to somebody's idea. If yours is close, check it against the market before you build: the Full Check names the rivals, the prices and the gaps.