I’ve been finding in-class "debates" one of the better ways to teach NeuroAI, since so much of it's developing intuitions about the various approaches to studying the brain, and more importantly why. For today’s class, I picked a paper (@dyamins & @JamesJDiCarlo's "Using goal-driven deep learning models to understand sensory cortex") & topic (“interpretability vs. predictivity”). The students had these roles: 1. Neuro/AI Scientist of the Past: Evaluates the work in the context of what was valued -10-20 years before 2. Neuro/AI Scientist of the Future: Evaluates the work from the perspective of the future (10+ years after). What parts of it will last and continue to impact, if any? 3. Pro: Argue for the “yes” side of the given prompt. (Group of 4) 4. Con: Argue against the given prompt. (Group of 4) 5. Jury: rest of class, who comments on the arguments given by each side. 6. New Experiments: This person suggests new experiments and future directions. The debate was super fun, and the students had really valuable insights on either side. There was also a rebuttal phase from both camps. In this case, throughout the course of the debate, it became clear that predictivity should guide the way to understanding, as it's been hard to go the other way and get something that predicts phenomena in the brain. Students made historical analogies to physics about how the deeper structure was revealed after the empirics gathering stage of that field (e.g. Kepler & Newton, Faraday & Maxwell, steam engine & thermodynamics).

Kai Cenat’s $70k AI humanoid robot just tried running away from the AMP house because it kept getting kicked and bullied by Kai, Agent & Fanum 😭😭