Two Camps
Sam Altman believes AGI could arrive within 5-10 years. Yann LeCun argues autoregressive LLMs are not the path to AGI.
Current Bottlenecks
- Reasoning: LLMs still struggle with complex reasoning
- World Models: Lack of physical world understanding
- Continuous Learning: Can’t update knowledge post-deployment
- Hallucination: Factual reliability remains a major issue — see LLM Hallucination Research
My Assessment
AGI likely needs 10-20 more years. But “useful enough AI” has already arrived, transforming knowledge work today.