Why neural foundation models work, and what they might—and might not—teach us about the brain

These models can partly generalize across species, brain regions and tasks, suggesting that a set of machine-learnable rules govern neural population activity. But will we be able to understand them?