Show HN: I built a web tool to see and edit what an AI thinks before it answers

DETAILED EXPLANATION

An independent developer built a free public tool that lets users see and edit what an open source language model is thinking mid-generation, based on Anthropic's recent paper on verbalizable representations forming a shared workspace inside language models. The tool uses a technique the developer calls a Jacobian Lens to visualize concepts lighting up in the model's middle layers before it commits to a word, and lets users manually insert or suppress concepts to change the output in real time.

Testing across models found the technique worked well on Qwen and Llama architectures but showed little benefit on Pythia models, regardless of size.