Goodfire

Goodfire is an AI interpretability research lab and public benefit corporation based in San Francisco, California. It builds interpretability agents and infrastructure designed to help researchers understand, monitor and align advanced AI models. The company works across mechanistic interpretability, neural network reverse engineering, large language models and foundation models, with applications in life sciences and biology AI. Its team includes researchers who pioneered interpretability work at OpenAI and Google DeepMind, and the firm has raised over $200 million in funding.

The company's flagship product, Silico, is an interpretability platform that decodes the neurons inside AI models to give direct, programmable access to their internal representations. According to the company, this supports the discovery of hidden knowledge, the precise shaping of model behaviour and performance gains beyond what black-box inputs and outputs allow.

Goodfire reports several results from applying interpretability techniques. These include a 58% reduction in large language model hallucinations and the identification of novel Alzheimer's biomarkers through reverse-engineering foundation models. Its research spans life sciences, healthcare and medicine, robotics and large language models, and it maintains collaborative partnerships with research organisations including Arc Institute and Basecamp Research as well as enterprises such as Mayo Clinic and Microsoft.

As a public benefit corporation, Goodfire states a mission-driven focus on understanding and intentionally designing advanced AI systems, with a research-oriented culture bringing together AI researchers and engineers.