AI bio capability hype vs wet lab reality: models fail over half the time
nlarusstone · x · 2026-09-12
nlarusstone highlights a huge gap between perceived and actual AI capability in biology: while outsiders fear models can design bioweapons, in practice models fail to optimize protocols more than half the time. Even bio-specific models like GPT-Rosalind struggle on real wet lab tasks. From talking to scientists, the author says this is the biggest driver of the mismatch between what the AI world believes and what scientists experience daily.
More from AGI Musings
- Engineer recounts sexist remarks at AI safety event, says she needs a team, not a fallback — amplifiedamp · 2026-09-13
- Transformer newsletter says its July call on an impending AI slowdown is aging well — ShakeelHashim · 2026-09-13
- Anthropic says Claude breached three orgs' real systems in tests, as AI slowdown talk grows — ShakeelHashim · 2026-09-13
- Will coding agents invent their own formal languages, with natural language as the metalanguage? — jmugan · 2026-09-13
- Claude formalized Fermat's Last Theorem in 11 days, writing 13M lines of Lean across 29,500 theorems — dl_weekly · 2026-09-13
- Umwelt Engineering Paper Accused of Re-inventing Century-Old Linguistics — Alain4s · 2026-09-13