Leading Goodfire’s $50M Series A to Interpret How AI Models Think

Computer science professor and founder of modern reinforcement learning Rich Sutton said in his essay “The Bitter Lesson” about AI, “Breakthrough progress eventually arrives by an opposing approach based on scaling computation by search and learning.” The breakneck progress of AI model development in the last three years is a testament to that lesson. As models, be it large language models or diffusion models, get bigger and better, they become more and more opaque. A black box. As these models get deployed in the world, they make critical decisions that impact all of our lives. Decisions which we frankly don’t understand. Would you be comfortable being in a plane where you don’t know how the autopilot system behaves? I wouldn’t.

When we first met with the Goodfire team, that was precisely the problem they’d set out to solve. They brought together a set of the strongest AI researchers and engineers to answer the question: Can we figure out why AI models behave the way they do? Can we steer that behavior? Goodfire is a company focused on core mechanistic interpretability research—the relatively nascent science of reverse engineering neural networks and using those insights to steer how all AI models behave. Our thesis was simple: In the future, AI models will be omnipresent, and we will not be comfortable with letting black boxes govern our lives. That is why I’m thrilled to announce that Menlo Ventures is leading Goodfire’s $50M Series A. Goodfire enables “brain surgery” on neural networks that was previously impossible.

We first partnered with Goodfire at the Seed with Lightspeed Venture Partners as one of the first investments from our Anthology Fund, which we co-lead with Anthropic. This round is personally meaningful to me for two reasons: It is the first company to “graduate” into a full Series A from the Anthology Fund, and it is my first board seat. As a show of conviction, our partners at Anthropic are joining us in this round, making their first direct investment into another company. Also participating are B Capital, Wing Ventures, and many others. 

There are few teams who are capable of a mission as lofty as Goodfire’s. CEO Eric Ho is a tenured founder who scaled his last company to north of $10M in ARR and is joined by his former head of AI Dan Balsam as a co-founder. Their team includes some of the best researchers in the field of interpretability, including co-founder Tom McGrath, Ph.D., who co-founded the interpretability team at DeepMind; Lee Sharkey, who pioneered the use of sparse autoencoders in language models and co-founded Apollo Research; and Nick Cammarata, who started the interpretability team at OpenAI alongside now Anthropic co-founder Chris Olah. Suffice to say, this is the best team to solve this problem.

Goodfire isn’t just a research lab with a mission; it’s made key strides in productizing and commercializing the understanding of models. Its platform, known as Ember, decodes the neurons inside of an AI model to give direct, programmable access to its internal thoughts. Ember unlocks entirely new ways to apply, train, and align AI models, allowing users to discover new knowledge hidden in their model, precisely shape its behaviors, and improve its performance. They’re the only lab to be able to interpret the recent DeepSeek R1 model. It’s not just text models; it applies to all kinds of AI models. Patrick Hsu, co-founder of Arc Institute—one of Goodfire’s earliest collaborators—who developed Evo 2, a DNA foundation model, with Goodfire, said, “Their interpretability tools have enabled us to extract novel biological concepts that are accelerating our scientific discovery process.”

“As AI capabilities advance, our ability to understand these systems must keep pace. Our investment in Goodfire reflects our belief that mechanistic interpretability is among the best bets to help us transform black-box neural networks into understandable, steerable systems—a critical foundation for the responsible development of powerful AI.”

—Dario Amodei, Co-Founder & CEO of Anthropic

We at Menlo Ventures recognize that core AI innovation happens when the best researchers meet world-class engineers to work on the most critical problems, and Goodfire is a testament to that belief. You can follow Goodfire on X to keep up with their latest developments!