DigestAI news desk
Research updated

OpenDiscoveryTrace: New Dataset Reveals AI Scientist Workflows

Researchers have released OpenDiscoveryTrace, a dataset of over 558 complete trajectories from AI scientists working on various scientific tasks. This dataset captures the reasoning process rather than just final outputs, providing valuable insights for auditing methodologies and diagnosing failures. The study includes seven models across different domains like drug discovery, materials science,…

1 source primary source

Key points

  • OpenDiscoveryTrace includes over 558 trajectories from seven models
  • Claude Opus 4.6 produces significantly more errors than GPT-5.4
  • Dataset covers diverse scientific tasks including drug discovery
Read the original at arXiv cs.AI · by Aayam Bansal, Keertan Balaji primary source Open source ↗
Topics · follow one to build your own front page
arXivOpenDiscoveryTraceGPT-5.4Claude Opus 4.6Gemini 3.1 ProQwen2.5-7BMistral-7B-v0.3Phi-3.5-mini

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.

Comments

via GitHub Discussions

More in Research

All →

Related stories