Season 2 · Episode 650
Can AI Solve Physics Problems It Never Learned?
Explore how Gemini 3.0’s Deep Think mode shifts AI from "fast" reflexes to "deliberate" reasoning to solve complex quantum physics problems.
My Weird Prompts · Daniel Rosehill
February 16, 202633m 58s
Audio is streamed directly from the publisher (dts.podtrac.com) as published in their RSS feed. Play Podcasts does not host this file. Rights-holders can request removal through the copyright & takedown page.
Show Notes
In this episode, Herman and Corn dive into the mechanics of Gemini 3.0 Pro’s new Deep Think mode and the fundamental shift from "System 1" pattern matching to "System 2" deliberate reasoning. They explore how models now use internal "scratchpads," Process-based Reward Models, and Monte Carlo Tree Search to solve problems that once seemed impossible, such as novel proofs in quantum physics. From the technical "sign problem" to the wild possibility of giving an AI a full week of compute to solve a single problem, this episode pulls back the curtain on the next frontier of artificial intelligence. It is a fascinating look at how "thinking longer" might be more important than "training bigger" in the quest for true machine intelligence.