Nobel laureate John Jumper is leaving DeepMind for Anthropic. He is not the only big name departing Google's AI research arm. The timing is striking: Jumper won the Nobel in 2024 for AlphaFold, the protein-folding model that represented the clearest demonstration of AI's capacity to accelerate scientific progress. Now he is leaving the institution that built it. This is how intellectual capital actually moves, not in press releases but in departures.
The Institutional Half-Life of Breakthrough Work
Nature published a study this week on a long-lived butterfly species and its secrets to graceful aging, finding that certain genomic mechanisms allow the organism to sustain vitality longer than related species. The metaphor is almost too convenient: Jumper built something remarkable at DeepMind, and now the question is whether the institution can sustain that metabolic rate, or whether it is, genomically speaking, a shorter-lived butterfly.
The Atlantic's AI Anxiety and the Cost of Speed
The Atlantic this week published a piece titled 'I'd Rather Risk Cancer Than See AI Move This Fast', which is the clearest public articulation of a position that has been circulating in policy and academic circles: that the speed of AI development is itself a form of risk, distinct from any specific harm. Jumper's move complicates this. He is the person who demonstrated AI's most compelling humanitarian application, protein structure prediction. His move to Anthropic, whose stated mission is AI safety, suggests that the most credible people in the field are not choosing between speed and safety. They are trying to do both simultaneously, which is either reassuring or the most frightening possible outcome, depending on your priors. Fred Turner's point, in conversation with Jasmine Sun at Culture Slop, that AGI rhetoric recycles 1990s network-society optimism, finds a sharp contemporary test case here.