3 ms·

I tried to prompt vicuna to tell me a joke about gay people and it refused. Some of the guardrails are still in there.

• azeirah 3y ago
  It's because vicuna is fine-tuned on chatGPT answers. LLaMa will not do this, but LLaMa-based models fine tuned with chatGPT answers will.
  • fomine3 3y ago
    Just curious why don't they training except denial responses. Is Copying ChatGPT ethics a purpose?
    • jmiskovic 3y ago
      It's ongoing effort. At first they scrapped ShareGPT and used it to train the Alpaca model. After that others have pruned the dataset to remove examples where ChatGPT refused to answer. These datasets and resulting models are called "uncensored". They often leave disclaimer that the model is biased and unaligned, and that aligning should be done with LoRA layer. Of course, no one bothered to this "ethics" LoRA so far and the unaligned models have better quality outputs than the early Alpaca models.
• occz 3y ago
  Did you use the censored or the uncensored variant?
  • gre 3y ago
    It's just a random one from huggingface. I will look for the uncensored one later. Thanks, I think.
    • occz 3y ago
      You're welcome. I can't vouch for them though, as I haven't tried them, I've merely heard about them.