Show HN: I built a web tool to see and edit what an AI thinks before it answers
Summary
Earthpilot's Lucid instrument reads a language model's internal representations (J-space) in the browser using a Jacobian lens to reveal concepts entering the workspace before the model answers. It runs on small open models and exports session slices for sharing, offering a hands-on look at how prompts map to internal reasoning. The project emphasizes interpretability and education with no account or install required.