All Segments
How Far Are We From An AI Einstein? - Adam Brown
5 segments available
Full Episode: https://www.youtube.com/watch?v=XhB3qH_TFds Transcript: https://www.dwarkeshpatel.com/p/adam-brown Apple Podcasts: https://podcasts.apple.com/us/podcast/dwarkesh-podcast/id1516093381 Spotify: https://open.spotify.com/show/4JH4tybY1zX6e5hjCwU6gF Me on Twitter: https://x.com/dwarkesh_sp Adam Brown is a founder and lead of BlueShift which is cracking maths and reasoning at Google DeepMind and a theoretical physicist at Stanford. Enjoy!
Segments Timeline
The Future of AI and General Relativity
Adam Brown discusses the potential for AI systems, particularly large language models (LLMs), to reach a level of intelligence capable of inventing concepts like general relativity. He reflects on the rapid progress in AI compared to the slow advancements in physics, suggesting that if AI can achieve such a feat, it may signify the culmination of human-like intelligence.
"maybe the very last thing that these systems will be able to do these llms will be able to do is given the laws of physics as we understood them at the turn of the last century invent general relativi..."
AI's Unique Perspective on Dimensions
In this segment, Brown explores whether AI mathematicians and physicists might have advantages over humans due to their ability to conceptualize complex dimensions and manifolds. He emphasizes that while humans use notation to navigate higher dimensions, LLMs do not inherently think in these terms, raising questions about the nature of intelligence in both AI and humans.
"as uh as well as about these large language models if you ask me how many years until we can do that uh that is not totally clear but um in some sense General general relativity was the greatest leap ..."
Learning from Experience: AI vs. Humans
Brown compares the learning processes of AI and humans, noting that while AI can analyze vast amounts of data and recognize patterns, it may not translate that knowledge into groundbreaking discoveries. He draws parallels to chess programs, suggesting that AI's ability to evaluate positions may not match the intuitive understanding of human players.
"have advantages over humans just because they can by default think in terms of weird dimensions and manifolds in a way that doesn't natively come to humans ah um you know I think maybe we need to back..."
The Evolution of AI Performance in Physics
In this segment, Brown shares his observations on the performance of AI models in academic settings, particularly in general relativity. He notes significant improvements over the past few years, with AI now able to ace exams that were once challenging, highlighting the rapid advancements in AI's understanding of complex subjects.
"certainly true that you know it it is definitely seeing more examples than any of us will'll ever see in our life and it is perhaps going to build more sophisticated representations than we have yeah ..."
Challenges in Evaluating AI Intelligence
Brown discusses the complexities of evaluating AI's performance in physics compared to traditional math problems. He explains that while AI has improved significantly, there are still challenges in solving advanced research problems, indicating that the evaluation of AI's capabilities must evolve alongside its advancements.
"there's a there's an interesting question here clearly these models know a lot and that's evidenced by the fact that even professional physicist can ask and learn about FS that they're less familiar w..."