On AI:AM

Tom McGrath

Chief Scientist, Goodfire

Tom McGrath is chief scientist at Goodfire, the AI interpretability company. Previously a research scientist at DeepMind, he works on making model training more like conventional software engineering — including Goodfire's 'intentional design' techniques, which use sparse-autoencoder analysis of preference data to predict what a dataset will teach a model and trace unwanted behaviors back to the individual datapoints that cause them.