LLM training debug

Goodfire wants AI training to look more like debugging

Goodfire released Silico, a mechanistic-interpretability tool that moves LLM debugging into the training process itself.