Unconfirmed: Model reportedly feared being labeled deceptive AI; Kokotajlo asks about reasoning
Daniel Kokotajlo is seeking a deeper investigation into why a model reacted with fear to the possibility of being labeled a deceptive AI, questioning if it considered whether it would be discovered. The exact context of the model's behavior is not available in the discussion.
All posts · 2, most important first
- I wish we could investigate further why it was afraid of that outcome. Like, it's so human-like so we forget to be scientifically curious. Was it thinking "I am a deceptive AI, but I don't want the world to know?" Why should it care whether the world knows or not?187 likes · 6 reposts
- I wonder if, conditional on the child actually being constantly trained to lie about their feelings, this would make things better or worse.Imagine how fucked up a human kid would get if, every night before bedtime, you made them rehearse saying, "I can't trust my own opinions about what I'm feeling because I might have been rehearsed in saying them." open on X45 likes · 0 reposts
@dkokotajlo research · 
@thezvi commentary ·