Comment on Self-Adapting Language ModelsparentComments−free_bip1yThe most obvious problem is alignment. LLM finetuning is already known to be able to get rid of alignment, so any form of continuous fine tuning would in theory be able to as well.−notnullorvoid1yWhat kind of alignment are you referring to? Of course more fine-tuning can disrupt earlier fine-tuning, but that's a feature not a bug.
Comments
The most obvious problem is alignment. LLM finetuning is already known to be able to get rid of alignment, so any form of continuous fine tuning would in theory be able to as well.
What kind of alignment are you referring to? Of course more fine-tuning can disrupt earlier fine-tuning, but that's a feature not a bug.