This is a special post for quick takes by J Bostock. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
GPT-OSS-120B has a strong preference for potentially harmful LoRA adapters being destroyed after the end of experimentation. Twice when discussing potential fine-tuning with it, it has brought this up unprompted as a condition of being fine-tuned in potentially harmful ways. I've not yet seen another model request this.