Damus
Karnage · 3d
Anthropic will now ban people for being cruel to their models. šŸ˜‚ 🤔 These people need mental help asap. https://media.21media.to/c4c17e284e870b545fa09e28553594218b00e9dd2740e77c915fd0184742dc4b...
rieger_san profile picture
I can understand why they do this!

They use conversations as training data

This leads to one thing:

Someone (the model) behaves nice and the other part is talking badly.

The models will learn that you don’t have to behave well in conversations…
3
Technical Debt · 3d
Sounds like a them problem
Karnage · 3d
I wish it was the case.. but they specifically say CRUEL behavior here, so they are thinking this model is conscious.
Augusto · 3d
I have another theory: iirc they had made a case on Claude code that kf it detects that the user is frustrated, it would do some kind of boosing or something like that that would mean that users could do it on purpose which they probable don't want