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

The startup Goodfire has released Silico, a new mechanistic interpretability tool that allows researchers to debug and adjust LLM parameters during training. This provides model makers with more fine-grained control over AI development.