Gyorgy Buzsaki: Brain Rhythms, Neural Syntax, and the Emergence of Cognition from Action 

Wow...what an insightful podcast!

Loving this. I studied SPWs over 10 years ago as an undergrad and I have a newfound interest in them because of the parallels to ML. SPWs are sort of the OG backpropagation and sleep is the training cycle that is required for information synthesis. Messing with SPWs during sleep can render a rat memoryless, as if the training never happened. Everything is just sequences all the way down, so curious how the computer scientists stumbled on LLMs which are sequential learning models.

Also really curious about dopamine's affect in reinforcement learning and how SPWs maybe actually lead to action. A huge problem in ML right now is how to make agents "creative" with reasoning capability. This is seen as sort of the next step in AI. I just took a break from this video to go get a burrito, which is about 5 miles away. At some point, my brain decided that the learned reward from a burrito would be higher than continuing to watch the video, and something drove my body to get up, get my keys, get in the car, and go to the burrito place. I surely have the path mapped out in my place cells, since I've been to this burrito place hundreds of times and I likely have a high reward associated with that place. Surely SPWs fired rapidly to pre-plan the trip, but what drove the brain to fire the SPW complex in the first place? Was it the learned reward? How does reinforcement learning relate to this? Did my physical hunger trigger my brain to fabricate the highest probability sequence in my mind that would lead to satiation? Surely there are hundreds of food options I'm aware of, but maybe the probability of planning out this particular burrito place was higher because I had thought about it a few times earlier in the day.