4 ms·
Uncensored finetunes aren't the same thing, that's taking a model that's already been lobotomised and trying to teach it that wrongthink is okay - rehabilitatio
by amenhotep 2y ago
Uncensored finetunes aren't the same thing, that's taking a model that's already been lobotomised and trying to teach it that wrongthink is okay - rehabilitation of the injury. OpenAI's uncensored model would be a model that had never been injured at all.
I also am not convinced by the argument but that is a poor reason against.
- staticman2 2y agoI'm talking about taking the Llama 3 base model and finetuning it with a dataset that doesn't include refusals, not whatever you mean by "taking a model that's already been lobotomized". It's interesting that you weren't convinced by the above argument but still repeated the edgelord term "lobotomized" in your reply.
- errantspark 2y agoThe claim is that llama is "lobotomized" because it was trained with safety in mind. You can't untrain that by finetuning. For what it's worth the non-instruct llama generally seems better at reasoning than instruct llama which i think is a point in support of OP.
- staticman2 2y agoBetter at reasoning based on benchmarks or what?