# AI Is Suddenly Surprisingly Good At Physics Source: https://www.youtube.com/watch?v=njNrTUMC6qE Recap page: https://rapidrecap.app/video/njNrTUMC6qE Generated: 2025-11-16T16:34:24.714+00:00 --- ## Quick Overview The video concludes that while Large Language Models (LLMs) like GPT-4 and Claude are improving in logic and math, they still lack the necessary ability to reason, validate, or produce truly novel scientific breakthroughs, making the bottleneck for progress the ability to properly value and validate scientific output rather than just generating more intelligence. The speaker highlights recent claims of AI making physics discoveries, referencing papers from startups like Extensity AI and Sakana AI, but ultimately expresses skepticism about true scientific reasoning capabilities in current LLMs, evidenced by anecdotal reports of failures and the need for human oversight. **Key Points:** - The main bottleneck in scientific progress is the ability to properly value and validate scientific output, not a lack of intelligence, according to researcher Bojan Tunguz. - Startups like Extensity AI claim to use AI (GPT-4/Codex) with neuro-symbolic software to analyze gravitational wave data for quantum gravity evidence, which the speaker finds impressive. - Google's AI co-scientist, built with Gemini 2.0, is being used to generate novel hypotheses and accelerate scientific discoveries, though the speaker notes they ran a trial and received no response. - Harvard professor David Sinclair claims his lab uses a novel AI system to make non-intuitive scientific discoveries and write up full PhD-level papers without human intervention. - A tweet from Asher suggests that claims of GPT-5 making novel research discoveries (happening twice in one week) are strange, implying potential overstatement or lack of independent validation. - The speaker notes that while LLMs are good at logic and math (as confirmed by Ekin Dogus Cubuk), they do not yet possess genuine reasoning capability, which is crucial for complex physics research. - The video concludes by showing an example of a custom star map product from Under Lucky Stars, suggesting that AI is currently better at generating personalized, creative products than fundamental scientific leaps. ![Screenshot at 0:04: The title card 'The AI Physicist' appears next to an image of a microchip engraved with quantum mechanics equations, visually setting the theme of AI intersecting with fundamental physics research.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-00-04.png) **Context:** The video, presented by Sabine Hossenfelder, discusses the rapidly evolving capabilities of Large Language Models (LLMs) in scientific research, specifically physics, contrasting claims of AI-driven breakthroughs with the speaker's inherent skepticism. The discussion centers on recent high-profile announcements from AI companies like Extensity AI and Google, as well as claims from prominent academics like David Sinclair, regarding AI's ability to generate novel scientific hypotheses and complete research papers autonomously. The host evaluates these claims against the fundamental limitations of current AI technology, particularly regarding true reasoning and validation. ## Detailed Analysis Sabine Hossenfelder reviews recent claims that AI is becoming surprisingly capable in physics research, beginning by discussing a paper from Extensity AI which uses AI and gravitational wave data to test quantum gravity, noting that she found their work impressive despite initial failures with GPT-4. She then mentions Google's AI co-scientist project, which aims to accelerate discoveries, but notes that her own application to their tester program went unanswered. The video contrasts this with claims from Bojan Tunguz, who argues that the main bottleneck in science is not intelligence but the ability to properly validate output, pointing to theoretical physics as full of unverified nonsense. Furthermore, David Sinclair of Harvard claims his lab is using an AI system that generates non-intuitive discoveries and writes full PhD-level papers autonomously. Hossenfelder counters these claims by citing anecdotal evidence, such as a tweet mentioning two unverified GPT-5 physics discoveries in one week, and by referencing Ekin Dogus Cubuk's point that while LLMs excel at logic and math, they lack true reasoning ability necessary for physics. The video finishes by contrasting these lofty scientific claims with commercial applications, showing a star map product from Under Lucky Stars, implying that current AI excels more at creative, personalized tasks than fundamental scientific work. ### AI in Physics Claims - Extensity AI uses AI/gravitational waves to test quantum gravity - Google introduces Gemini 2.0-based AI co-scientist for hypothesis generation - David Sinclair claims AI writes PhD-level papers without human intervention ### Skepticism and Bottlenecks - Bojan Tunguz states validation, not intelligence, is the scientific bottleneck - LLMs are good at logic/math (Ekin Dogus Cubuk) but lack true reasoning ability ### Anecdotal Evidence & Trials - Speaker's trial for Google's AI co-scientist yielded no response - Tweet notes two unverified GPT-5 physics discoveries in one week ### Commercial AI Application - Video pivots to showcasing 'Under Lucky Stars' app creating personalized star maps, suggesting AI excels at creative gifts rather than foundational science. ![Screenshot at 0:01: Title screen showing Sabine Hossenfelder and a chip with physics equations, introducing the theme of AI in physics.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-00-01.png) ![Screenshot at 0:13: A cartoon robot labeled 'ChatGPT' appears next to the physics chip, symbolizing the AI tools being discussed.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-00-13.png) ![Screenshot at 0:34: Sabine Hossenfelder gestures while discussing the AI scientist concept.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-00-34.png) ![Screenshot at 0:41: A man points to code on a projection screen, representing the programming/simulation aspect of AI research.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-00-41.png) ![Screenshot at 1:04: The Extensity AI website displays their mission: 'AI Research Automation' using neuro-symbolic research.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-01-04.png) ![Screenshot at 1:12: A blog post graphic shows an artistic rendering of a black hole, referencing Extensity AI's work on black hole cores using AI.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-01-12.png) ![Screenshot at 2:15: Ekin Dogus Cubuk explains that current LLMs are good at logic and math but lack true reasoning.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-02-15.png) ![Screenshot at 2:46: A tweet highlights that Period Labs raised $300M and plans to develop the first generation of AI scientists.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-02-46.png) ![Screenshot at 3:35: An image of money bags and coins appears, illustrating the discussion about monetizing AI scientific output.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-03-35.png) ![Screenshot at 4:14: A digital clock graphic appears, symbolizing the concept of saving time through AI research.](https://ss.rapidrecap.app/screens/njNrTUMC6qE/00-04-14.png)