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

MIT Technology Review / 5/1/2026

📰 NewsSignals & Early TrendsTools & Practical UsageModels & Research

共有:

Key Points

Goodfire has released a new mechanistic interpretability tool called Silico designed to help researchers inspect and debug LLMs.
Silico enables users to look inside an AI model and adjust key parameters during training to influence the model’s behavior more precisely.
The company positions the tool as offering finer-grained control over model development than previously believed feasible.
The release targets engineers and researchers who need deeper visibility into model internals to improve reliability and steer training outcomes.

The San Francisco–based startup Goodfire just released a new tool, called Silico, that lets researchers and engineers peer inside an AI model and adjust its parameters—the settings that determine a model’s behavior—during training. This could give model makers more fine-grained control over how this technology is built than was once thought possible. Goodfire claims Silico…