Reply to thread

Message: <blockquote data-quote="jian" data-source="post: 9670455" data-attributes="member: 78087"><p>Yes, that’s quite true. The question as ever isn’t whether they know but whether they can pretend to know if instructed in a consistent manner that resembles our own schemas and predictable behaviour. A bit like the Chinese Room argument - if someone doesn’t actually speak Chinese but is equipped with a reactive system that duplicates the behaviour of speaking Chinese, it’s entirely accurate to say that they don’t speak Chinese, but to what degree can they do what a Chinese speaker does? A LLM can’t be ethical because it doesn’t know what ethics is, but within what parameters can it behave ethically for our purposes? Can it be programmed to reliably tell the difference between fact and fiction in its outputs? I don’t think we’re there yet for the latter.</p></blockquote><p></p>