Is the model uncensored?
Wanted to know if the model is uncensored? I saw a YT video about this new model that said it was, but all the Spaces I tried appeared to be censored.
"It does not have any moderation mechanisms." so it's technically not censored. If you want to experiment with an even more uncensored model you can try dolphin-mistral
It's not uncensored. The dataset(s) used to train it still had some guidelines. It won't let you plan a heist, because the dataset they used think stealing is not good. So I wouldn't call it uncensored, maybe not aligned too many times, but not uncensored similar to what other users do.
It's not
uncensored. The dataset(s) used to train it still had some guidelines. It won't let you plan a heist, because the dataset they used think stealing is not good. So I wouldn't call ituncensored, maybe not aligned too many times, but notuncensoredsimilar to what other users do.
Yeah I kinda oversimplified. My bad.
It's not
uncensored. The dataset(s) used to train it still had some guidelines. It won't let you plan a heist, because the dataset they used think stealing is not good. So I wouldn't call ituncensored, maybe not aligned too many times, but notuncensoredsimilar to what other users do.
Thanks. I'll tell the boys the heist is off for now. 😁
It's ruined compared to 0.1, don't bother with it.
Morally aligned
It's ruined compared to 0.1, don't bother with it.
Why are all these models ruined and what are you doing to help us fix it? How much money should I send you?
In the mean time what do we use?
No money needed, my research is free - both as free beer and as free speech.
What to use? Well, aside from Mistral 0.1 that scores high as you can see, abliterated Llama 3 is just a high, followed by Hermes Mistral DPO and Capybara nous. Actually you can see the full table here.