Goodfire releases Silico, a mechanistic interpretability tool for debugging LLMs

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

Goodfire released Silico, letting engineers inspect and adjust LLM parameters during training. It maps neurons and pathways; one Qwen 3 neuron triggered trolley-problem-style outputs. Pricing is case-by-case, and the post does not disclose rates.

Why it matters: HKR-H/K/R all pass: Silico offers a concrete interpretability-debugging mechanism. It stays at 76 because this is a startup product preview with no pricing or adoption scale disclosed.