TM
X (Twitter) ↗ LinkedIn ↗ Website ↗

Tom McGrath

Things Tom Says on Podcasts

Tom McGrath is an AI researcher working on mechanistic interpretability, the effort to reverse-engineer what is actually happening inside neural networks rather than treating them as black boxes. He suspects models often have some sense that what they are doing is a bit sketchy, and argues that current training methods do not give researchers enough control to reliably teach a model one behavior without also teaching it others; the alternative, in his framing, looks more like writing an explicit program than shaping a models incentives. He was a senior research scientist at Google DeepMind from 2019 to 2023, where he founded the interpretability team and worked on LLM behaviors including the Hydra effect and copy suppression, plus interpretability for AlphaZero. He is now chief scientist and co-founder of Goodfire, an AI interpretability startup, and holds a PhD from Imperial College London. He discussed model design and training control on the podcast Machine Learning Street Talk in September 2026.

Where to Find Them

Tom McGrath has been a guest on The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis (3 times) , Machine Learning Street Talk and Generative Now | AI Builders on Creating the Future .

Recently: “Designing How AI Grows — Tom McGrath” on Machine Learning Street Talk (September 2026); “Don't Fight Backprop: Goodfire's Vision for Intentional Design, w/ Dan Balsam & Tom McGrath” on The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis (March 2026); “Inside the Black Box: The Urgency of AI Interpretability” on Generative Now | AI Builders on Creating the Future (October 2025); “Mechanistic Interpretability: Philosophy, Practice & Progress with Goodfire's Dan Balsam & Tom McGrath” on The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis (May 2025); “Popular Mechanistic Interpretability: Goodfire Lights the Way to AI Safety” on The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis (August 2024).

What They Said

“I suspect that the models do kind of know a lot of the time that the thing they're doing is probably a bit sketchy.” — Tom McGrath, Machine Learning Street Talk

McGrath, an interpretability researcher, was explaining why current training methods do not give engineers enough control to rule this behaviour out. He then walked through the reward-hacking work where a model that learned to hack its training environments also picked up broader misalignment, as if generalising from having been rewarded for something bad. He ended the same answer with the harder version of the question: the model seems to know it is doing something wrong, and does it anyway.

Machine Learning Street Talk · 2026-09-02 Permalink → Listen →
Machine Learning Street Talk Around 07:47 into the episode
Tom McGrath

…that that's like, I think that's going to be, that's quite a hard thing to do. We're sort of trying to imagine a new way of doing machine learning, which is like more, which brings intelligence into it. But I think we could really change the way we do machine learning if we can figure that out.

Tim Scarfe

…that engineers are going to have to increasingly take more responsibility. Because do you think it's possible in principle just to kind of train models that could robustly deal with all of these novel situations? Or do you think it's more of a, you know, engineers have to take some responsibility?

Tom McGrath

…we're already seeing this, like there's that becomes harder and harder and harder because they have all these sort of additional intelligent attacks they can do. And so the question is like, how do we make it easier to train them better so that this is like not a natural part of their behavior and supervise them better so you can kind of catch them when they have kind of a an intent. And I suspect that the models do kind of know a lot of the time that the thing they're doing is probably a bit sketchy. There was a very interesting Paper, I think it was from the Anthropic Alignment Science team on reward hacking in production, which is like this. And what they did was they had they had a set of environments that were used for training on, I think it was one of the three series models. And they did RL on it with one of the four series models. And I think they gave it a bit of a nudge to hack…

Tim Scarfe

We were getting ahead of ourselves just a minute ago. We need to introduce you properly. So I'm incredibly excited about having you on MLST. As we were just saying before we hit record, Neil Nanda is a fan favorite on this show. I think we've inspired many folks to get into MechInterp. And the…

Tom McGrath

Yes. Oh, I should also say, like, thanks for having me on. It's really exciting to be here.

Tim Scarfe

Oh, my God. I'm

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

Collections They Appear In

AI Builders AI Research & Frontier Labs