Show HN: LLM Thought Visualization

Category: ai-ml

Tags: llm-interpretability, visualization-tool, model-internals

Score: 7.3/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 6)

Subtext is a real-time visualization tool that uses Jacobian lensing to expose the internal verbal workspace of a language model during live conversation, revealing thoughts like forming verdicts and planning ahead before output. It combines research from Anthropic on J-space with a practical interactive demo, making model interpretability more accessible and engaging.

Target audience: ai researchers, ml engineers, interpretability enthusiasts

Repository: https://github.com/ninjahawk/Subtext · HTML · NOASSERTION · 1 stars

View on Hacker News