Narrow finetuning can produce broadly misaligned LLMs
emergent-misalignment.comIs anybody pointing on the fact that "alignment" is brain-washing?
Is teaching a child? Is talking with your friend? Is punishing a criminal?
Or, more pointedly, what about training the model in the first place? Why do you pretend that AI are somehow "people" with a "natural tendency" we're overriding?
discussed 2 days ago: https://news.ycombinator.com/item?id=43176553