Sam Altman on AGI, OpenAI, and the Future of AI — Lex Fridman Podcast #367

The March 2023 Lex Fridman conversation with Sam Altman (#367, 1hr 47min) was the defining public interview of the ChatGPT era — conducted at the peak of public excitement about GPT-4 and Altman's most extended public treatment of AGI timelines. Altman's key claims: he genuinely believes AGI is possible within years not decades; the most important question is whether AI development can be steered toward broad benefit rather than concentrated power; and OpenAI's governance structure (capped-profit, board accountability) is designed to preserve the mission even against OpenAI's own commercial interests. The conversation is notable for Altman's uncertainty about timelines — he refuses to give specific years and emphasizes that the field is unpredictable — but his overall tone is urgency without panic. He discusses the challenge of alignment: current RLHF methods work for current models but may not scale to more capable systems, and OpenAI is betting that the research problems will be solved in time. His description of "the vibes" of model evaluations — qualitative sense of capability jumps before benchmarks confirm them — revealed how much of frontier AI assessment relies on tacit knowledge unavailable in papers. Altman's willingness to discuss the board structure and governance in detail — unusual for a tech CEO — reflected the November 2023 crisis that hadn't yet happened but was already latent in OpenAI's organizational design.