Full threadshikon7·It seems straightforward that finetuning an LLM just for usefulness will reduce alignment (and the other way around).View on HN