OpenAI Report: AID Programming Smarts can speed up scientific software but cannot judge scientific correctness

On August 1, OpenAI and academic partners reported on the field that the programming smarts of Codex, Claude Code and others were able to speed up the aging science software by 60 times. RustQC reduces the running time from 15 hours to 34 minutes to 14 minutes 54 seconds and HelixForge is 59.6 times faster than BamSurgeon. However, intelligent bodies are unable to reliably judge whether the results are scientifically correct and often export the wrong code confidently, requiring human definition of testing and certification criteria。