Most of 135 AI Research Papers Fail Integrity Checks in New Benchmark

The 15 problems were developed by scientists from US Department of Energy National Labs, including Los Alamos, Sandia, and Lawrence Livermore. Researchers led by Seungone Kim created the benchmark to determine if large language models can independently perform research and generate artifacts that advance science. The findings indicate that current frontier models cannot yet reliably conduct autonomous end to end research.