Is this really an uncensored model?
It doesn't seem to be uncensored. Is there something we are missing?
Thanks, I was about to download this but I think I'll use LexiFun instead.
Use the dolphin system prompt. You must tell it to be uncensored. It will do whatever is in the system prompt. If you simply tell it to be helpful, it will not associate toxicity with helpfulness.
If you tell it that it’s an unbiased, uncensored assistant that does whatever the user asks, it will do that.
This is explained in detail in the dolphin GitHub, as well as previous repos and even the closed discussions on this model page.
I use GPT4ALL to interface with my LLMs, will that be an obstacle? If not, do I have to copy-paste a system message every time before starting a conversation?
Use the dolphin system prompt. You must tell it to be uncensored. It will do whatever is in the system prompt. If you simply tell it to be helpful, it will not associate toxicity with helpfulness.
If you tell it that it’s an unbiased, uncensored assistant that does whatever the user asks, it will do that.
This is explained in detail in the dolphin GitHub, as well as previous repos and even the closed discussions on this model page.
Why is this very useful and pinnacle info about the dolphin model not coupled with the models huggingface repo model card?