A preprint proposes a framework for models that help set their own problems
single source· 1 articles · confidence: medium · first seen 2026-09-13 20:00 UTC
What this means for you
Nothing to act on today. No weights, no API, and no evaluation results are reported for either named system — this is an argument for how discovery work should be structured, plus a code link. Worth tracking if you build automated experimentation pipelines; otherwise there is nothing to migrate to or test.
A preprint posted to arXiv on 13 September proposes "discovery foundation models" — general-purpose systems meant to take part in creating new problems and explanations, not just solving ones people specify. The authors list seven coupled capabilities, including problem formulation, representation construction and evidence-grounded revision, and name two instantiations: Zetema, a software system with verification and experimental gating, and GALILEO, which the paper calls a therapeutic-discovery system pairing automated reasoning with robotic and hands-on lab experiments. No benchmark scores, evaluation dates or results from either system are reported. Code is linked on GitHub.
Key facts
- ·The paper was posted to arXiv on 13 September 2026 under identifier 2609.15973. source
- ·It proposes "Discovery Foundation Models" as general-purpose systems operating over what the authors call a revisable research state. source
- ·It names seven coupled capabilities, spanning problem discovery, formulation, representation construction, hypothesis formation, intervention, evidence-grounded revision and continual discovery improvement. source
- ·Two instantiations are given: Zetema, and GALILEO, which the paper describes as a therapeutic-discovery system combining Dry-Lab reasoning with robotic and hands-on wet-lab experimentation. source
- ·The paper reports no benchmark scores, evaluation dates or results from either system. source
- ·Code is linked at github.com/Gen-Verse/DFM-Plans. source
What the sources say
- Hugging Face Daily Papers (research) — Position paper naming two systems and a code repository; no experimental results or evaluations reported for either.
Sources
The original reporting. Follow these — they did the work.
- Hugging Face Daily PapersDiscovery Foundation Models: Toward Open-Ended Discovery Intelligence2026-09-13