All Segments
If an LLM solves this then we'll probably have AGI – Francois Chollet
3 segments available
Full Episode: https://youtu.be/UakqL6Pj9xo Transcript: https://www.dwarkeshpatel.com/p/francois-chollet Apple Podcasts: https://podcasts.apple.com/us/podcast/francois-chollet-mike-knoop-llms-wont-lead-to-agi-%241/id1516093381?i=1000658672649 Spotify: https://open.spotify.com/episode/7bmeJQOvXGy4LYl6YoiYYP?si=obUSUEwjSA6tkB8EBcb18w Follow me on Twitter: https://x.com/dwarkesh_sp
Segments Timeline
The Path to AGI: Solving ARC
Francois Chollet discusses the implications of a multimodal model achieving 80% accuracy on the ARC benchmark. He explores the challenges of creating a perfect benchmark that cannot be anticipated and the potential for brute-forcing intelligence through task memorization. This segment delves into the significance of ARC in evaluating artificial general intelligence (AGI) and the limitations of current models.
"so suppose that it's the case that in a year a multimodal model can solve Arc let's say get 80% whatever the average human would get then AGI quite possibly yes I think if you if you start so honestly..."
Intelligence as Pathfinding
Chollet presents a metaphor for intelligence as a pathfinding algorithm in a future situation space, akin to navigating a foggy map in real-time strategy games. He emphasizes that intelligence relies on past experiences and the limitations of knowledge in anticipating future changes. This segment highlights the distinction between human learning and machine memorization, questioning the nature of intelligence in both contexts.
"know with enough scale you can always cheat if you can do this for every single thing that supposedly requires intelligence then what good is intelligence apparently you can just Brute Force intellige..."
Memorization vs. Learning in AI
In this segment, Chollet critiques the notion of memorization in both humans and AI, arguing that what we label as learning often involves memorizing skills and techniques. He discusses the implications of automating tasks through synthetic data and the economic impact of AI on remote work. Chollet raises questions about whether such automation still falls under the memorization regime, emphasizing the challenges posed by dynamic environments.
"simply memorizing every possible path every mapping from uh point A to point B uh you could you could solve the problem with pure memory but where the reason you cannot do that in real life is because..."