Why I'm Betting Against the AGI Hype
10 days ago
- #LLMs
- #AI Hype
- #AGI
- The author argues that current LLM-based approaches are useful tools but unlikely to achieve AGI due to fundamental architectural limitations.
- Key constraints include the separation of learning and inference, catastrophic forgetting worsening with scale, and energy inefficiency.
- The author highlights a category error in treating consciousness as an engineering problem rather than a complex phenomenon.
- Compound probability analysis suggests a low (1-15%) chance of AGI breakthroughs from scaled LLMs in the near future.
- The economic bubble around AGI is driven by narrative momentum and misaligned incentives rather than technical feasibility.
- Expert opinion is bimodal, with industry leaders optimistic and academic researchers skeptical about LLMs leading to AGI.
- The author bets against the AGI hype, emphasizing the need for fundamentally different approaches to intelligence.