LLM-Visualized explains GPT-2 with live tensors

This feels like a real teaching artifact, not an AI gimmick: it turns transformer internals into something you can inspect, which is far more memorable than another static diagram.

• –Real activations and attention scores make it more credible than toy explainers or marketing demos
• –KV-cache, prefill, and decode views are especially valuable because they explain a key inference trick most visualizations skip
• –Browser-first delivery makes it easy to use in classrooms, workshops, and blog posts without setup friction
• –GPT-2 Small keeps the demo approachable, but it also limits how far the project can scale as a general-purpose debugger
• –Early feedback already points to the main gap: newcomers need clearer narration and simpler onboarding