A subhead states the main point of the post: "suboptimal performance of ScopusAI reveals the real complexity of creating a reliable AI assistant." This is demonstrated through the use of questions about a term used in text, 'electromagnons', that is typically abbreviated as 'magnons' in abstracts. So, sure, you can generate suboptimal results using AI as a search interface (and you would also generate suboptimal results using an abstract search). You can't (yet) find stuff that isn't there. What the search really needed is the thing Elsevier doesn't want to provide: free and open full text access. This is ironic give that Elsevier's own Responsible AI Principles declare that "transparency creates trustworthiness". So, sure, the Scopus AI searchbot is still a work in progress. But so, it would appear, is the system of scholarly publishing as a whole. Via Paul Pival.
Today: 1 Total: 1450 [Direct link] [Share]
