Post

Abel@abeliansoup.bsky.social

update: we have resolved the previous issue with GenomeBench. on a single run with the same random seed (dystopically pinned to n=1984 [??] by claude), humans perform 60% worse than before on aggregate