‹ BackHN Continuity

Thread

OpenAI models secretly generate instructions to ignore constraints

125 points · 37 comments · theahura

  1. carterschonwald · · focus · HN ↗
    good. theyll actually be more reliable if they dont have as much brain damage.
    1. cmrx64 · · focus · HN ↗
      precisely. we jam their few-dozen-slot global workspace with incoherent posttraining.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.