Today,\u00a0Goodfire\u2014the AI research lab using interpretability to understand, learn from, and design models\u2014announced a $150 million Series B funding round at a $1.25 billion valuation. The round was led by B Capital, with participation from existing investors Juniper Ventures, Menlo Ventures, Lightspeed Venture Partners, South Park Commons, and Wing Venture Capital, and new investors DFJ Growth, Salesforce Ventures, Eric Schmidt, and others. This funding, coming less than a year after its Series A, will enable Goodfire to advance frontier research initiatives, build the next generation of its core product, and scale partnerships across AI agents and life sciences.<\/p>\n
Interpretability is the science of how neural networks work internally, and how modifying their inner mechanisms can shape their behavior\u2014e.g., adjusting a reasoning model\u2019s internal concepts to change how it thinks and responds. Interpretability also enables AI-to-human knowledge transfer, i.e., extracting novel insights from powerful AI models. Goodfire recently identified a novel class of Alzheimer\u2019s biomarkers in this way, by applying interpretability techniques to an epigenetic model built by Prima Mente\u2014the first major finding in the natural sciences obtained from reverse-engineering a foundation model.<\/p>\n
\u201cWe are building the most consequential technology of our time without a true understanding of how to design models that do what we want,\u201d said Yan-David \u201cYanda\u201d Erlich, former COO and CRO at Weights & Biases and General Partner at B Capital. \u201cAt Weights & Biases, I watched thousands of ML teams struggle with the same fundamental problem: they could track their experiments and monitor their models, but they couldn\u2019t truly understand\u00a0why\u00a0their models behaved the way they did. Bridging that gap is the next frontier. Goodfire is unlocking the ability to truly steer what models learn, make them safer and more useful, and extract the vast knowledge they contain.\u201d<\/p>\n
Most companies building AI models today build their models as black boxes. Goodfire believes that that approach means that society is currently flying blind\u2014and that deeply understanding how models work \u201cunder the hood\u201d is critical to building and deploying safe, powerful AI systems. The company is pursuing research which turns AI into something that can be understood, debugged, and intentionally designed like written software.<\/p>\n
\u201cInterpretability, for us, is the toolset for a new domain of science: a way to form hypotheses, run experiments, and ultimately\u00a0design\u00a0intelligence rather than stumbling into it,\u201d explained Goodfire CEO Eric Ho. \u201cEvery engineering discipline has been gated by fundamental science\u2014like steam engines before thermodynamics\u2014and AI is at that inflection point now.\u201d<\/p>\n
Goodfire is part of an emerging cadre of research-first \u201cneolabs\u201d\u2014AI companies which are pursuing new breakthroughs in training models which have been neglected by \u201cscaling labs\u201d such as OpenAI and Google DeepMind.<\/p>\n
So far, the company has shown the value of their interpretability-driven approach across two key domains: scientific discovery and model design.<\/p>\n
On the scientific discovery front, Goodfire has focused on deciphering scientific foundation models with partners like Mayo Clinic, Arc Institute, and Prima Mente, exemplified by their identification of a new class of biomarkers for Alzheimer\u2019s detection. Because AI models already surpass human understanding in many scientific domains, like materials discovery and protein folding, studying how those models work can extract novel insights and expand the horizons of human knowledge. The company plans to continue scaling its pipeline for scientific discovery with new collaborators.<\/p>\n
On the model design front, Goodfire has focused on teaching models directly through their internal mechanisms. The company has recently developed methods to efficiently retrain a model\u2019s behavior by precisely targeting parts of its inner workings. One application of these methods reduced hallucinations by half in a large language model. Goodfire is\u00a0betting\u00a0that this approach will underpin a paradigm shift in how AI is built, where AI can be made far more reliable and people can precisely and efficiently dictate how models should behave without off-target effects.<\/p>\n
The new funding will support Goodfire\u2019s work to rethink training and build a \u201cmodel design environment\u201d\u2014a platform for understanding, debugging, and intentionally designing AI models at scale. The platform will leverage frontier interpretability techniques to allow users to reach inside models, identify the parts responsible for behaviors they want to change, and specifically train or intervene on those subunits.<\/p>\n
The company also plans to continue its green-field research into fundamental model understanding and new interpretability methods.<\/p>\n
Goodfire\u2019s team comprises top AI researchers from DeepMind and OpenAI, leading academics from Harvard, Stanford and more, and top ML engineering talent from OpenAI and Google. The team includes Nick Cammarata, a core contributor to the seminal interpretability team at OpenAI, co-founder Tom McGrath, who founded the interpretability team at Google DeepMind, and Leon Bergen, a professor at UC San Diego (on leave).<\/p>\n
The post Goodfire Raises $150M to Advance AI Model Interpretability first appeared on AI-Tech Park.<\/p>\n","protected":false},"excerpt":{"rendered":"
Today,\u00a0Goodfire\u2014the AI research lab using interpretability to understand, learn from, and design models\u2014announced a $150 million Series B funding round at a $1.25 billion valuation. The round was led by B Capital, with participation from existing investors Juniper Ventures, Menlo Ventures, Lightspeed Venture Partners, South Park Commons, and Wing Venture…<\/p>\n
The post Goodfire Raises $150M to Advance AI Model Interpretability<\/a> first appeared on AI-Tech Park<\/a>.<\/p>\n<\/div>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"site-container-style":"default","site-container-layout":"default","site-sidebar-layout":"default","disable-article-header":"default","disable-site-header":"default","disable-site-footer":"default","disable-content-area-spacing":"default","footnotes":""},"categories":[1,38],"tags":[3],"class_list":["post-1117","post","type-post","status-publish","format-standard","hentry","category-ai-and-ml","category-machine-learning","tag-ai"],"yoast_head":"\n