Anshul Kundaje#1779
@InVitroFuture @p_maverick_b @baym These restrictions show the deciders have never set foot in a biolab. Even simple things often don't work, and engineering and *testing* a pathogen is very far from simple. It's not like vibe coding, guys.
Many users dismissed Anthropic's biosafety policy team as lacking practical understanding of AI bio tool restrictions and relying on metrics uncorrelated with reality.
@TomKnightSynBio @p_maverick_b @baym If Anthropic's biosafety policy team doesn't have a good understanding of the practical steps necessary to develop harmful new forms of infectious material, all the more reason for them not to trust themselves to write rules to distinguish such work from innocuous bio research
@jvarga92 @TomKnightSynBio @InVitroFuture @p_maverick_b @baym They鈥檙e indexing on metrics that have little correlation to reality, and conveniently ignore that most of biology today is not measurable.
@InVitroFuture @TomKnightSynBio @p_maverick_b @baym Then they should hire ppl that do know!
@TomKnightSynBio @InVitroFuture @p_maverick_b @baym And what about all the tech bros announcing every day that they came up with the newest shiny super binder design pipelines, then show high iptm/whatever values only without any experimental testing? Like as if it was not shown to be not useful for measuring binding 100s of times
Anshul Kundaje#1779@InVitroFuture @p_maverick_b @baym These restrictions show the deciders have never set foot in a biolab. Even simple things often don't work, and engineering and *testing* a pathogen is very far from simple. It's not like vibe coding, guys.