We need more empirical research on biological AI model safety by Noga Aharony

Epistemic sta­tus: We are quite con­fi­dent that Biolog­i­cal AI mod­els (BAIMs) safety re­quires fur­ther work, but un­cer­tain about its scale. The ap­par­ent gap may be re­futed by one ex­per­i­ment, filled by a few re­searchers work­ing for a year, or prove to be sub­stan­tial enough to call for an en­tire sub­field.Dis­claimer: this post has been writ­ten with a col­league, that due to her cur­rent job can’t post out of her own fo­rum accountWhat we’re look­ing for: please poke holes in thi...