Evaluation Argues That AI Agents Cannot Currently Conduct Open-Ended Research

CruxEvals presents an evaluation arguing that current AI agents cannot reliably conduct open-ended research. The available Hacker News record shows a score of 1 and one comment, but does not expose the evaluation methodology, task set, model coverage, metrics, or experimental results. The claim is therefore useful as a research-capability hypothesis, but its evidentiary strength cannot be assessed from the supplied metadata alone.