Yair Halberstadt proposes materials science and bioscience as the near-term capability tripwire to watch

LessWrong·low signalXBlueskyLinkedInCopy link

Posted to LessWrong on September 14, the argument is that the meaningful threshold is not whether models can do science but whether they can do it one-shot, without iteration, using the Navier-Stokes counterexample as the template for a discovery-shaped problem. Halberstadt names room-temperature superconductors and cancer cures as the watch items and treats a one-shot breakthrough as the signal for bioweapon design risk. He includes David Bellamy's counterargument from X that virus design needs iterative wet-lab work rather than pure reasoning, and the post offers no success rates or empirical data, so this is a framing proposal rather than a result.