Any system prompt degrades the quality of the model's results, even the suggested one. Moreover, the model only pretends to be less censored, but it is more censored in practice. Wasting compute again, huh?
· Sign up or log in to comment
