John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI
Dwarkesh Podcast
00:00
Coherence is the main next step towards AGI
Improving coherence in AI models for longer periods could lead to human-level interaction capabilities, making them nearly as good as interacting with a human colleague. However, besides coherence, current AI models also have weaknesses in critical thinking, understanding tasks, and paying attention. Merely enhancing coherence may not be sufficient for achieving Artificial General Intelligence (AGI), but it could be a significant advancement towards it. Planning for the potential emergence of AGI in the near future would be a reasonable approach.
Play episode from 15:59
Transcript
