Embedded AI evaluators, censorship concerns and pressure to slow development
One post argues that embedding evaluators in AI companies would create a censorship layer. A reply says companies face pressure to slow AI while also facing lawsuits alleging collusion over slowing down for safety.
TLDR
A critic calls the push to embed evaluators inside AI companies “Trust & Safety 2.0,” alleging that an unaccountable NGO layer would enable censorship and use safety concerns to pressure companies through the media. A reply describes conflicting demands: criticism that companies aren’t slowing AI enough, alongside lawsuits alleging that major AI companies colluded over slowing down for safety. The reply sums up that tension as “Damned if you do, damned if you don’t.”
Combined views
165.3K
1 Source, first seen ago