No way I'm watching that rn, but I admire your ambition.
I hope he gets around to making more of the LLM series soon. I might get motivated enough to rewatch the series. I'm pretty sure it was 3 brown one blue that describes the math about how more data is stored in neurons than the bits themselves, how that property emerges from math, and results in more complex contextual nuances that are not scientifically well understood. This is where much of model alignment behavior and conceptual abstractions are happening. It is this area that I find most fascinating to intuitively explore.