New benchmark tests whether AI can do real scientific discovery

Original: ProjectionBench: Evaluating Scientific Hypothesis Generation in LLMs Under Progressive Information Disclosure

Source: arxiv.org ↗