@deedydas - Hassabis and Mallaby Discuss

Open original

@deedydas on X At a San Francisco event, Demis Hassabis and Sebastian Mallaby covered AI's near-term risks, from OpenAI's finances to who controls powerful cyber tools. Hassabis also pitched Isomorphic Labs' drug discovery work and urged more careful, science-style decision-making on AI.

  • Mallaby put the chance of OpenAI going bankrupt within 18 months at 50%.
  • Hassabis said Isomorphic Labs needs six "AlphaFold" moments to shrink drug development from about 10 years to a few months.
  • Mallaby argued it isn't tenable for a private company to decide who gets access to frontier cyber defense tools, since China could match them in 6 to 12 months.
  • Hassabis rejected the 0% p(doom) view held by Marc Andreessen and Yann LeCun, calling it "crazy."

Demis Hassabis and Sebastian Mallaby spoke on stage in San Francisco, and the most attention-grabbing claim came from Mallaby: he put a 50% chance on OpenAI going bankrupt within 18 months. Hassabis, who runs Google DeepMind, praised Anthropic CEO Dario Amodei as the best of the other lab leaders. Mallaby also raised a policy concern about Anthropic's Claude Mythos, a model with frontier cyber defense capabilities. His point was that it isn't tenable for a private company to decide who gets access to that technology, especially when a rival like China could replicate it within six to twelve months. That is a pointed argument for government involvement in access decisions, not just corporate ones. On the science side, Hassabis put the emphasis on Isomorphic Labs, the drug discovery company spun out from DeepMind. He said AlphaFold, the protein structure prediction system that earned him a share of the 2024 Nobel Prize in Chemistry, is only one of the problems that need solving. He estimated the field needs six more "AlphaFold moments" to shrink drug development timelines from roughly ten years to a few months. On existential risk, he rejected the idea of putting firm probabilities on p(doom), but said the risk is non-zero. He singled out Marc Andreessen and Yann LeCun, who put it at zero, as wrong. He also said he was optimistic about AI sentiment in India, based on his time at the AI Summit hosted by Prime Minister Modi, where he found the public far less pessimistic than in parts of the West. Hassabis's longer-range view was more reflective. He described a post-scarcity future with an enormous amount of new science, but said the hard problems would be economic, namely how to share the proceeds equitably, and philosophical, requiring a new generation of thinkers. He framed his personal role as pushing for more rigorous, CERN-style collaboration and stress-testing of ideas, with social scientists and economists involved in decisions. For anyone early in their career, his practical advice was to immerse yourself in AI tools, since the publicly available versions run only about three to six months behind the frontier, and to look for applications in fields nobody has touched yet.

Ask about this article