This is why LLMs are terrible for summarizing a scientific literature. It will reproduce the current consensus.
But we move science forward by challenging the consensus, pointing out flaws and holes in what people think and especially with how they've conducted research or drawn inferences.
But we move science forward by challenging the consensus, pointing out flaws and holes in what people think and especially with how they've conducted research or drawn inferences.
Reposted from
Kevin M. Kruse
Even if we ignore the tendency to make things up completely, ChatGPT is horrible for historical research because it generates its "answers" by surveying the existing historical work out there, which means (1) the ideas are not novel in any way and (2) the writing is, by definition, "average"
Comments