“We found that 60 out of the 76 papers (79 percent) that claimed to outperform a standard numerical method had used a weak baseline, either because they hadn’t compared to more advanced numerical methods, or because they weren’t comparing them on an equal footing.”
A physicist checked AI-for-science papers that claimed huge speedups. 79% compared against weak baselines. The most cited physics-informed network paper left out its failures. Scientists adopt AI partly because it helps their careers.