Really disturbing if true - more evidence that we do not really understand what is going on inside of LLMs and that "traits" (including dangerous ones) can be passed from model to model invisibly. We need unbiased research/regulation. But we got Pedo Felon. www.theverge.com/ai-artificia...