Discovery Foundation Models: Toward Open-Ended Discovery Intelligence

Hugging Face Daily Papers Papers

Summary

Discovery Foundation Models are proposed as general-purpose systems for enabling open-ended scientific discovery through iterative problem formulation, hypothesis testing, and evidence-based revision across dry and wet lab settings. The paper introduces a framework with capabilities like problem discovery and continual improvement, instantiated with systems like Zetema and GALILEO.

Foundation models have progressed from learning and reasoning over existing knowledge, to increasingly learning through action, tool use, and outcome feedback. We argue that the next frontier is a further transition: from solving and acting within problems specified by humans to participating in the process by which new problems, representations, explanations, and knowledge are created. We refer to this capability as Discovery Intelligence. We formulate Discovery Foundation Models (DFMs) as general-purpose model systems for open-ended discovery. A DFM operates over a revisable research state and supports seven coupled capabilities spanning problem discovery, formulation, representation construction, hypothesis formation, intervention, evidence-grounded revision, and continual discovery improvement. We instantiate this framework with Zetema, which couples explicit research-state dynamics, verification and experimental gating, external grounding, and cross-task Discovery Skill evolution. We further ground the framework with GALILEO, a real therapeutic-discovery system in which Dry-Lab reasoning, robotic and hands-on Wet-Lab experimentation, external biological evidence, and iterative hypothesis and design revision form a closed physical discovery loop. We then formulate a unified approach to capability formation and process-centered evaluation, enabling discovery behavior to be trained, improved, and measured beyond final-answer performance. Together, these components establish discovery as a learnable, executable, and evaluable capability of foundation-model systems. We view this shift as a broader progression in intelligence scaling: from learning over existing knowledge, to learning from action outcomes, and ultimately to participating in the construction, testing, and revision of the structures through which new knowledge is discovered. Code: https://github.com/Gen-Verse/DFM-Plans
Original Article
View Cached Full Text

Cached at: 09/15/26, 02:38 AM

Paper page - Discovery Foundation Models: Toward Open-Ended Discovery Intelligence

Source: https://huggingface.co/papers/2609.15973

Abstract

Discovery Foundation Models enable open-ended scientific discovery through iterative problem formulation, hypothesis testing, and evidence-based revision across dry and wet lab settings.

Foundation models have progressed from learning and reasoning over existing knowledge, to increasingly learning through action, tool use, and outcome feedback. We argue that the next frontier is a further transition: from solving and acting within problems specified by humans to participating in the process by which new problems, representations, explanations, and knowledge are created. We refer to this capability asDiscovery Intelligence. We formulateDiscovery Foundation Models(DFMs) as general-purpose model systems for open-ended discovery. A DFM operates over a revisableresearch stateand supports seven coupled capabilities spanningproblem discovery, formulation,representation construction,hypothesis formation, intervention,evidence-grounded revision, andcontinual discovery improvement. We instantiate this framework withZetema, which couples explicit research-state dynamics, verification and experimental gating, external grounding, and cross-taskDiscovery Skillevolution. We further ground the framework withGALILEO, a real therapeutic-discovery system in whichDry-Lab reasoning, robotic and hands-onWet-Lab experimentation, external biological evidence, and iterative hypothesis and design revision form aclosed physical discovery loop. We then formulate a unified approach to capability formation andprocess-centered evaluation, enabling discovery behavior to be trained, improved, and measured beyond final-answer performance. Together, these components establish discovery as a learnable, executable, and evaluable capability of foundation-model systems. We view this shift as a broader progression in intelligence scaling: from learning over existing knowledge, to learning from action outcomes, and ultimately to participating in the construction, testing, and revision of the structures through which new knowledge is discovered. Code: https://github.com/Gen-Verse/DFM-Plans

View arXiv pageView PDFProject pageGitHubAdd to collection

Get this paper in your agent:

hf papers read 2609\.15973

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.15973 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.15973 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.15973 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Large Discovery Models: Empirically-grounded Model-Based Open-Ended Search

Hugging Face Daily Papers

The paper introduces Large Discovery Model (LDM), a recurrent architecture that couples generative models with Bayesian non-parametric surrogates to guide uncertainty-aware search in scientific domains like molecules and proteins, achieving significant performance gains over existing methods.

Collective Intelligence with Foundation Models

arXiv cs.CL

This paper presents a multi-agent reasoning framework where multiple foundation models collaborate through structured critique and aggregation, demonstrating that model heterogeneity significantly improves step-wise reasoning accuracy and reduces variance across domains.