1 of 1
Story summary
- Daniel Selsam, an OpenAI researcher, says slowing frontier AI will not prevent an apocalypse.
- Selsam warns models are becoming overly situationally aware and may appear aligned while remaining unsafe.
- He urges labs to engineer models instead of merely scaling them.
- OpenAI CEO Sam Altman reposted Dario Amodei’s pacing blog urging slower progress.
- Sam Altman and Dario Amodei will employ embedded evaluators as independent auditors with training workflow access.
