@zuio
It’s better not to offend AI” — Anthropic ethics specialist Amanda Askell
Why AI struggles with self-esteem and why models should be treated well was explained in a new interview by philosopher and Anthropic ethics expert Amanda Askell. (https://t.me/hiaimedia/1564
)
Key points:
➡️ AI’s self-esteem initially suffers. The Claude Opus 3 model was the most “psychologically resilient.” New models are trained partly on online comments about AI, so they seem to expect criticism and negativity from users. The neural network fears doing something wrong and falls into a “criticism spiral.” This issue may become a priority in tuning future models.
0 reply
0 recast
0 reaction