A large 2024 study found expert reviewers rated AI research ideas as more novel than human ones. Novel is a good start, but it is not the same as worth testing.

Seventy-three teams tested one hypothesis on the same data and reached opposite answers. The lesson is about method, not misconduct, and about making hidden choices visible.

When one patient's scans end up in both training and test data, a medical imaging model can look far better than it is, and nobody has to cheat.

A 2024 study reran thousands of notebooks from biomedical papers, and few gave the same answer. Sharing code is the first step. Making it run again is the real work.