AI + a16z The Best Way to Achieve AGI Is to Invent It
32 snips
Nov 4, 2024 
Pedro Domingos, a machine-learning pioneer and author, joins Martin Casado for an intriguing discussion on the quest for Artificial General Intelligence (AGI). They delve into the costs associated with scaling AI and the feasibility of reaching human-level intelligence without astronomical investments. The conversation also touches on the importance of embracing non-traditional perspectives to spark innovation, and Domingos shares insights from his satirical novel that critiques the intersection of AI and society, blending humor with deep philosophical questions.
AI Snips
Chapters
Books
Transcript
Episode notes
AI's Impact
- Current AI research sprints towards local optima, like making Transformers efficient.
- This won't achieve human-level intelligence, but even 10% progress significantly impacts the world.
Scaling Laws
- Scaling laws in AI, while seemingly promising now, are controversial and may not hold.
- Human brains aren't simply scaled-up ant brains, suggesting scaling alone won't achieve AGI.
LLMs as World Models
- LLMs are currently used as world models due to abundant text data, but this might be temporary.
- Pedro Domingos believes LLMs will likely persist as interfaces, generating text based on real knowledge and reasoning.
