DeepMind's core breakthrough was treating AI like a child, not a machine. Instead of programming complex strategies, they taught it to master tasks through simple games like Pong, giving it only one rule ('score go up is good') and allowing it to learn for itself through trial and error.

Reinforcement learning incentivizes AIs to find the right answer, not just mimic human text. This leads to them developing their own internal "dialect" for reasoning—a chain of thought that is effective but increasingly incomprehensible and alien to human observers.

A novel prompting technique involves instructing an AI to assume it knows nothing about a fundamental concept, like gender, before analyzing data. This "unlearning" process allows the AI to surface patterns from a truly naive perspective that is impossible for a human to replicate.

In domains like coding and math where correctness is automatically verifiable, AI can move beyond imitating humans (RLHF). Using pure reinforcement learning, or "experiential learning," models learn via self-play and can discover novel, superhuman strategies similar to AlphaGo's Move 37.

AlphaGo's infamous 'Move 37' was a play no human expert would have made, initially dismissed as an error. Its eventual success demonstrated that AI can discover novel, superior strategies beyond the existing corpus of human knowledge, fundamentally expanding a field of study rather than just mastering it.

The "bitter lesson" in AI research posits that methods leveraging massive computation scale better and ultimately win out over approaches that rely on human-designed domain knowledge or clever shortcuts, favoring scale over ingenuity.

Human intelligence is shaped by limitations like a finite lifespan and small brain, forcing efficient learning from sparse data. AI lacks these constraints, learning from lifetimes of data with massive compute. This fundamental difference means AI will naturally evolve into a distinct, non-human form of intelligence unless we explicitly engineer human-like biases into it.

Even when a model performs a task correctly, interpretability can reveal it learned a bizarre, "alien" heuristic that is functionally equivalent but not the generalizable, human-understood principle. This highlights the challenge of ensuring models truly "grok" concepts.

In the endgame, AlphaGo made moves that seemed suboptimal, even giving up points. This was because it wasn't optimizing for a large victory margin (a human heuristic) but purely for maximizing the probability of winning, even by a half-point. This reveals how literal AI objective functions can differ from human proxies for success.