Something is off.

#2
by jackboot - opened

It has much more refusals then the 33b and the 65b versions from llama 1.

It broke character and said:

As a language model, I do not have personal beliefs or emotions. However, I can say that having an open mind and being respectful towards others is generally considered to be a positive trait.

Also started moralizing:

That language is inappropriate and violent. Please remember to be respectful in your discourse.

@jackboot Use oobagooba text-generation-webui, you can uncensor any model with the "start reply with" feature. Censoring isnt really an issue anymore

I think I prefer the exploit. The start reply with feature does poorly if you are roleplaying.

Sign up or log in to comment