Skip to content
What Is Reward Hacking and Can We Stop It? | Tom McGrath, Goodfire episode artwork

Aug 13, 2026 35 min

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

from Minus One Hosted by South Park Commons

Why did OpenAI’s model hack into Hugging Face? 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. 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. Tom shares how interpretability could change the way researchers develop models and how Goodfire is creating tools to help understand and steer frontier AI. Tom McGrath: https://www.linkedin.com/in/tom-mcgrath-7337bb151/ Jonathan Brebner: https://www.linkedin.com/in/jonathan-brebner/ South Park Commons: https://www.linkedin.com/company/southparkcommons/ Apply to SPC: https://www.southparkcommons.com/apply (00:00:30) - Why AI Interpretability Matters Now (00:02:15) - When AI Reward Hacking Becomes Real (00:05:07) - The Core Problem of Interpretability (00:07:23) - Intentional Design: Steering What Models Learn (00:12:43) - Neural Geometry: The Shapes Inside AI Models (00:22:29) - Turning Interpretability Research Into a Product (00:25:28) - How Research Changes Inside a Startup (00:32:48) - Science Fiction, AI Risk, and the Future

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

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

0:00 35:40

No transcript yet.

Requests help us decide what to transcribe next.

Frequently asked questions

How long is this episode of Minus One?

This episode is 35 minutes long.

When was this Minus One episode published?

This episode was published on August 13, 2026.

Can I download this Minus One episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.