Import AI 475: Swarm scaling; Google DeepMind watermarks biology; and the AI science economy

Import AI covers swarm scaling, a US AI-rules poll, DeepMind's SynthID Bio, and lab-automation benchmarks.

Import AI's roundup leads with Toby Ord's argument that AI swarms are a form of inference scaling: they spend more tokens overall but finish faster through parallel agents, with diminishing returns he likens to economists' stepping-on-toes effect. A Center for Shared AI Prosperity poll of 2,498 Americans found 61% consider voluntary industry AI commitments insufficient, and 54% want government-enforced rules. Google DeepMind introduced SynthID Bio, watermarking methods for amino-acid sequences and predicted protein structures; wet-lab tests on VEGF-A, the SARS-CoV-2 spike RBD, and PD-L1 matched unwatermarked hit rate and binding. C5R's SciUniverse benchmark covers 92 tasks in 17 families for AI systems operating partly automated labs.