This startup’s new mechanistic interpretability tool lets you debug LLMs

# A Startup Just Released a Tool That Lets You See Inside AI Models and Fix Problems A San Francisco company called Goodfire has created a new tool that allows AI developers to look directly at how an AI model works while it's being trained, and then tweak the settings to change how the AI behaves. Think of it like being able to open up the hood of a car while it's running and adjust individual parts—except here, it's for artificial intelligence. This could make AI systems more reliable and easier to control, since developers can now spot and fix issues before the model is finished.