What Is Reward Hacking and Can We Stop It? | Tom McGrath, Goodfire

<p>Why did OpenAI’s model hack into Hugging Face?</p><p>Tom McGrath, co-founder and Chief Scientist at Goodfire, joins South Park Commons Partner Jonathan Brebner to explore one of AI’s biggest challenges: interpretability.</p><p>Using OpenAI’s recent reward hacking incident as a starting point, they discuss why today’s most advanced models are still difficult to understand and why gaining visibility into how they work could be essential for building safer, more reliable AI systems.</p><p>Tom shares how interpretability could change the way researchers develop models and how Goodfire is creating tools to help understand and steer frontier AI.</p><ol><li>Tom McGrath: <a href="https://www.linkedin.com/in/tom-mcgrath-7337bb151/" rel="ugc noopener noreferrer" target="_blank">https://www.linkedin.com/in/tom-mcgrath-7337bb151/</a> </li><li>Jonathan Brebner: <a href="https://www.linkedin.com/in/jonathan-brebner/" rel="ugc noopener noreferrer" target="_blank">https://www.linkedin.com/in/jonathan-brebner/</a> </li><li>South Park Commons: <a href="https://www.linkedin.com/company/southparkcommons/" rel="ugc noopener noreferrer" target="_blank">https://www.linkedin.com/company/southparkcommons/</a></li></ol><p>Apply to SPC:<a href="https://www.southparkcommons.com/apply" rel="ugc noopener noreferrer" target="_blank"> https://www.southparkcommons.com/apply</a></p><p><br /><br />(00:00:30) - Why AI Interpretability Matters Now</p><p>(00:02:15) - When AI Reward Hacking Becomes Real</p><p>(00:05:07) - The Core Problem of Interpretability</p><p>(00:07:23) - Intentional Design: Steering What Models Learn</p><p>(00:12:43) - Neural Geometry: The Shapes Inside AI Models</p><p>(00:22:29) - Turning Interpretability Research Into a Product</p><p>(00:25:28) - How Research Changes Inside a Startup</p><p>(00:32:48) - Science Fiction, AI Risk, and the Future</p>