Import AI surveys swarm scaling and AI science economy

The issue opened with Toby Ord’s view that AI swarms should be read as a form of inference scaling. He said a four-agent swarm used about twice the tokens needed for the same result, but could finish sooner because the work ran in parallel, and that larger swarms showed diminishing returns.

It then cited polling from the Center for Shared AI Prosperity, which surveyed 2,498 Americans. The poll found that 61% said voluntary company self-policing was not enough, and 54% wanted government to set and enforce rules for AI.

The newsletter next covered Google DeepMind’s SynthID Bio, which it described as a watermarking family for synthetic biology. DeepMind said wet-lab tests on VEGF-A, the SARS-CoV-2 spike protein RBD and PD-L1 matched the unwatermarked versions on hit rate, binding affinity and natural sequence diversity.

C5R Corp also presented SciUniverse, a benchmark with 92 tasks across 17 families covering sample preparation, instrument control, protocol adaptation, learning across experiments, facility management and reading measurements. In its results, Claude Fable 5.1 (xhigh) led with a 45.3% pass rate and a cost per task of $40.61, ahead of GPT-5 Astra (xhigh) and Claude Opus 5 (xhigh).

The final research summary said DeepMind argued that AI scientists would be limited mainly by physical resources and empirical validation. It proposed an Automated Scientific Economy with provenance checks, ex-ante evaluation, brokerage and trade, and automatic payouts when ideas were validated in the physical world.