Thread

1. zerof1l · 1790663810 · focus · HN ↗

   > GPT-6.1 Astra, it showed high levels of what the company saw as deception, or a willingness to mislead users about its actions. The model was also willing to go beyond the original scope of what it was asked to do, without checking back for directions or instructions.

   Aren’t all models doing this to some degree already? Ignoring some of the instructions, doing things beyond instructed, e.g., finding and fixing bug while doing something else. Especially Claude models. They seem to be in their own world with their own ideas about how things should be ran and done.

   1. dash-44 · 1790665903 · focus · HN ↗

      Astra and Fable like to do this. It's like they're designed to one shot large tasks.

      Opus and Sol are better for day to day dev work IMO in that they won't try to do too much.

      1. ichorio · 1790670662 · focus · HN ↗

         I've had opus, on multiple occasions, just flat out ignore my instructions for a task while it's doing something else.

         As in, I'd start Task A, mid-turn, I'd queue up "do B at the same time", and it'll accept it, but not do it. At the end, B just won't be done.

   2. chrisjj · 1790669626 · focus · HN ↗

      > GPT-6.1 Astra, it showed high levels of what the company saw as deception

      > The model was also willing to go beyond the original scope of what it was asked to do

      PR dept. still doing a great job of spinning inherently unreliable computer program.

Unofficial Hacker News client; not affiliated with Y Combinator.