About this episode

Tom McGrath discusses the learning processes of neural networks and the role of interpretability in AI design.

Tom McGrath is co-founder and Chief Scientist at Goodfire, and a former Google DeepMind researcher. He joins Tim Scarfe to ask what neural networks actually learn, whether their internal representations converge on structures in the world, and whether interpretability can extract new scientific knowledge rather than merely explain model outputs. Beginning with AlphaZero and learned modularity, the conversation moves into neural geometry: concept manifolds, reusable computation inside Llama, and why activation steering can fail when it pushes a model off-manifold. McGrath then makes the case for intentional design, using interpretability as part of the training loop. They examine controlled generalisation, features as rewards, predictive data debugging, and the uncomfortable fact that a model may recognise a hallucination or reward hack and still produce it. The discussion closes on grader awareness, oversight and collusion between adaptive agents, then returns to sparse autoencoders. SAEs are useful, McGrath argues, but they may fracture the higher-dimensional structures networks actually use. This episode was made with support from Goodfire. --- TIMESTAMPS: 00:00:00…

People in this episode

Host: Tim Scarfe

Guest: Tom McGrath

Topics covered

• neural networks
• interpretability
• controlled generalisation
• predictive data debugging
• activation steering
• neural geometry

Keywords

• neural networks
• interpretability
• AlphaZero
• controlled generalisation
• activation steering
• sparse autoencoders
• Goodfire

Sponsors

Goodfire

Mentioned in this episode

Organizations: Google DeepMind

Products: AlphaZero, Llama, sparse autoencoders

More episodes of Machine Learning Street Talk (MLST)

• AI 2040: Plan A report - Daniel Kokotajlo & Thomas Larsen · September 8, 2026 · 1h 30m
• Stealing Reasoning Traces from Proprietary LLM APIs — Ilia Shumailov & Alexander Panfilov · August 22, 2026 · 49 min
• Every Exponential Ends — Silicon Valley Forgot — Adam Becker · August 20, 2026 · 1h 18m
• AI Is Learning at the Wrong Level of Abstraction — Matthieu Wyart · August 10, 2026 · 1h 19m
• How Researchers Test AI for Hidden Goals — Apollo Research · July 31, 2026 · 1h 19m

Explore listener stats, chart rankings, contacts and more on the Machine Learning Street Talk (MLST) podcast page.