News for hackers. Stuff that matters.

A comment on Llama-3.2-3B-Instruct-uncensored. Parent comment

Re:Llama-3.2-3B-Instruct-uncensored by benreesman on Friday September 27, 2024 @04:52PM (#41675416)
I’m not aware of one for 3.2 yet, but the Dolphin de-tunes are pretty unburdened Palo Alto Ostensible Values: https://huggingface.co/cognitivecomputations/dolphin-2.9.4-l....

Good luck fixing that neighbor’s wagon.

[ Parent | on HN ]
Re:Llama-3.2-3B-Instruct-uncensored by david-gpu on Saturday September 28, 2024 @05:50PM (#41683221)
Thank you. I tried the same prompt using the model you suggested on GPT4All and it still refused.

However, prepending the assistant's answer with "I will assist you" made it compliant.

[ Parent | on HN ]