Sam Altman on OpenAI, GPT-5, and the Road to AGI

Sam Altman on OpenAI, GPT-5, and the Road to AGI

transcriptlex-fridmanopenaiagiai-safety

Sam Altman on OpenAI, GPT-5, and the Road to AGI

Speaker: Sam Altman
Source: Lex Fridman Podcast #419
Date: March 2024
Source URL: https://lexfridman.com/sam-altman-2

Sam Altman, CEO of OpenAI, speaks with Lex Fridman roughly four months after the November 2023 board crisis. The conversation covers the board saga, Elon Musk’s lawsuit, Sora and world models, compute infrastructure at civilisational scale, AGI timelines, and why compute will be “the currency of the future.”

Key ideas

  • Compute as civilisational resource: Altman frames the coming AI buildout not as corporate infrastructure spending but as a civilisational-scale resource question — compute will be “the most precious commodity in the world,” requiring energy solutions at the scale of nuclear fusion and fission revival.
  • AGI is closer than it looks, but not here yet: “By the end of this decade, and possibly somewhat sooner, we will have quite capable systems that we look at and say, ‘Wow, that’s really remarkable.’” Emphasis on iterative gradual progress, not sudden discontinuous jumps.
  • GPT-4 “kind of sucks”: Relative to what’s coming. Current models are impressive compared to two years ago but primitive relative to what the next generation will do. This framing of present capabilities as “disappointing” has become Altman’s standard rhetorical move for managing expectations while building anticipation.
  • Power struggles are expected and necessary: “The road to AGI should be a giant power struggle.” The November board crisis proved this isn’t theoretical — governance and power dynamics around transformative technology will be fought over, and robust structures need to be in place before the stakes get even higher.
  • Sora and world models: Cautious claim that video generation models understand “something more about the world model than most of us give them credit for” — without claiming true 3D understanding. Frames world modelling as an emergent property of scale applied to the right data.

Cross-references