Yes, while I personally believe this is a problem to solve, not an insurmountable obstacle, I cannot ignore this kind of harm. I do think this will be outweighed by all the instances where they improve decision making, dig up obscure information, correct errors, etc. I can’t see any fundamental reason a near future AI system has to hallucinate anything, and when we also have people complaining in thread that LLMs are dumping out song lyrics verbatim…well, it’s an amusing contrast to say the least.