AI Uncensored: a fine tune of Llama 405B – no more PC nonsense • Hacker News Board

AI Uncensored: a fine tune of Llama 405B – no more PC nonsense

#1

Post by saroyas » Thu, Aug 22, 2024, 2:38 PM UTC

AI Uncensored: a fine tune of Llama 405B – no more PC nonsense
aiuncensored.info

Re: AI Uncensored: a fine tune of Llama 405B – no more PC nonsense

#2

Post by lawlessone » Thu, Aug 22, 2024, 2:41 PM UTC

Is the title here editorialized? I don't see that second part on the page?

Re: AI Uncensored: a fine tune of Llama 405B – no more PC nonsense

#3

Post by yorwba » Thu, Aug 22, 2024, 3:01 PM UTC

lawlessone wrote: post #2

Is the title here editorialized? I don't see that second part on the page?

It's editorializing when someone other than the author does it. This is just an ad: https://news.ycombinator.com/from?site=aiuncensored.info

Re: AI Uncensored: a fine tune of Llama 405B – no more PC nonsense

#4

Post by nickpsecurity » Thu, Aug 22, 2024, 3:10 PM UTC

Do they have the training data for the fine-tune somewhere? Or blog posts about how they did it?

More here for those interested:

https://news.ycombinator.com/item?id=35946060

Re: AI Uncensored: a fine tune of Llama 405B – no more PC nonsense

#5

Post by lawlessone » Thu, Aug 22, 2024, 3:17 PM UTC

yorwba wrote: post #3

lawlessone wrote: post #2

Is the title here editorialized? I don't see that second part on the page?

It's editorializing when someone other than the author does it. This is just an ad: https://news.ycombinator.com/from?site=aiuncensored.info

It seems like a misunderstanding of why models are censored. I don't think companies censor their models to adhere to some nebulous "PC nonsense".

They do it because having their model advocate for something bad could potentially make them liable if someone acts on something bad they goaded a model into saying.

e.g. that nutcase that tried to kill the queen of England.