Another fascinating AI paper, this one trying to address the lack of factual reliability from LLMs. These researchers take inspiration from the law by having one LLM "cross-examine" another. Interesting approach. But I wonder how either even can identify factual claims to interrogate.
(And -- somewhat off topic -- I wonder what happens when LLMs end up in extended dialogue; do they start making up their own short-cut language?)
Follow
@jeffjarvis I don't understand why people are even studying this topic. It's like someone doing a study on which type of shoe works best as a hammer.
