AI & Tech
AI Cannot Learn Real World Skills Like Building Businesses Without Sample Efficiency
Dwarkesh Patel Podcast
The next big breakthrough will be AIs learning on the job
"How do we train an AI to get really good at building a business from scratch? How about winning court cases? Or having a profitable day of trading in the markets? Or helping a candidate win an election? The rollout here requires interacting with the real world, and you can't recreate it from just within a data center."
The speaker argues that critical real-world skills like entrepreneurship, litigation, trading, and political strategy cannot be trained through current reinforcement learning methods because they require months or years of real-world interaction that cannot be simulated in parallel rollouts. This represents a fundamental limitation in the path to AGI that scaling compute alone cannot solve.
From this episode
Dwarkesh Patel Podcast