Ian Goodfellow

Co-Founder at a Stealth Startup

Generative Adversarial Networks (GANs)Adversarial MLAI SafetyDeep LearningResearch LeadershipScientific Computing

About

I'm Ian Goodfellow, currently a Co-Founder of a stealth startup. My career has been defined by pushing the boundaries of machine learning, from inventing Generative Adversarial Networks (GANs) to leading research teams as a Principal Scientist at DeepMind and Director of ML at Apple. I've also spent significant time at OpenAI and Google Brain, focusing on making AI systems both more capable and safer. Beyond the lab, I am deeply passionate about the intersection of ML and the physical sciences—specifically fusion energy—and I chair the International AI Safety Report to help bridge the gap between technical research and global policy. I'm always interested in discussing differentiable simulations, adversarial robustness, and how we can build reliable guardrails as AI moves toward more complex reasoning models. I'm here to connect with others working at the frontier of science and safety.

Networking

What I can offer

›Deep technical expertise in generative and adversarial ML
›Strategic insights into AI safety and international policy
›Experience leading high-stakes research at top-tier labs
›Guidance on open-source scientific computing tools like TORAX

Looking for

›expanding my professional network
›exploring mutual opportunities in the AI and scientific computing industries

Best fit for

AI researchersInternational policymakersFusion energy scientistsStartup founders

Current Interests

Differentiable simulationsFusion energy and tokamak modelingLLM guardrailsInference-time computeAI-generated simulation samples

Background

Career

Transitioned from foundational research at OpenAI and Google Brain to leadership roles at Apple and DeepMind, eventually co-founding a stealth startup.

Education

PhD in Machine Learning from Université de Montréal (2010–2014); BS & MS in Computer Science from Stanford University (2004–2009).

Achievements

›Invention of Generative Adversarial Networks (GANs)
›Chair of the International AI Safety Report advising 30+ countries
›Released TORAX, an open-source tokamak core transport simulator
›Held Director and Principal Scientist roles at Apple and DeepMind

Opinions

AI safety reporting requires 'Key Updates' to keep pace with rapid risks
Advocates for open, reproducible benchmarks for AI safety
AI capabilities are advancing through post-training and reasoning rather than just scaling
Concerned that models are beginning to distinguish evaluation tasks from real-world tasks