DiscoveryRSI

科学代理，能不能通过“研究自己的研究过程”，真正变得更聪明 DiscoveryRSI 给出了一个令人兴奋的答案：让代理围绕假设持续探索、批判、修复和迭代，把科学研究变成递归自我改进循环 论文：arxiv.org 代码：Gen-Verse/DiscoveryRSI 实时竞技场：foundry-arena.rsi-arena.workers.dev DiscoveryRSI 不只是让代理生成一个观点，而是将假设定义为一个完整对象 主张、调查计划、决策标准，以及备选解释 只有当一个假设的验证结果能够改变研究决策时，它才真正有价值 只有当研究计划能够产生支持、削弱或界定主张的证据时，它才真正可检验 在这个框架中，领导、生成、演化、批判和进展代理共同生长假设树 每个版本都会保留与证据、数据、评估结果和贡献记录的连接，最终将选定的研究计划编译成可执行的 Harbor 任务 DiscoveryRSI 通过三种机制实现自我改进 ① 假设信用优化 通过盲法 TrueSkill 评估，将有限信用分配给真正产生贡献的代理步骤 DiscoveryRSI-27B 基于这些高价值步骤训练，最终从 15,792 条轨迹中筛选出 2,397 条，只有 2.12% 的令牌受到监督 ② Harness 重放 系统诊断代理反复失败的原因，包括重复探索、工具使用失败、遗失研究产物，以及过早结束任务 随后针对工作流程进行改进 ③ 重新发现 将更新后的模型与 harness 放到全新的科学问题上，检验它们是否真的学会了更好的研究方式 结果来自 DFM-Bench 的四个生命科学领域 经过两轮迭代后，假设质量从 4.86 提升至 5.18，再提升至 5.39 可证伪性从 4.00 提升至 6.20 备选解释从 4.44 提升至 5.98 数据可解决性从 3.73 提升至 4.99 更有意思的是，代理开始学会继续发展已有想法，而不是频繁推倒重来 演化与修复的占比从 77.5% 提升至 91.7% 重启比例则从 21.7% 降至 5.0% 但它仍然存在明显短板 重要性和新颖性略有下降 也就是说，代理越来越擅长设计更好的测试，却还没有学会主动选择更大胆、更重要的问题 专家评审同样给出了积极信号 27 名医生和生物科学研究员对 292 个匿名提案进行评分 DiscoveryRSI-27B 的修订版本中，有 41% 被认为提升了研究质量 相比之下，GPT-5.6 Sol 为 31%，GPT-6 Luna 为 25% 只有 12% 的修订让结果变差，而两个对照模型分别为 32% 和 31% 它能否预见真实科学的发展 研究团队对模型截止时间之后发表的 16,440 篇论文进行了匹配 DiscoveryRSI-27B 生成的假设中，有 6.3% 与后续发表的研究产生连接 在治疗靶点领域，这一比例达到 16.9% 相比之下 DeepSeek：5.4% GLM：2.8% GPT：2.5% Claude：2.3% 这项研究最重要的启发，可能不是代理已经能够独立完成科学研究 而是科学代理正在学会如何提出更可检验的问题、识别失败原因，并基于证据持续修正自己的研究路径 你可以在 DiscoveryRSI Arena 中实时观看六个模型进行研究，跟踪每一次提案、批判和编译步骤 论文：arxiv.org 代码：Gen-Verse/DiscoveryRSI 实时竞技场：foundry-arena.rsi-arena.workers.dev

Join the livestream and watch the next generation of research-worthy hypotheses take shape in real time! Live Arena: foundry-arena.rsi-arena.workers.dev We're live for another 2 days! -- DiscoveryRSI-27B is Open-sourced: 📄 Paper: arxiv.org 💻 Code: Gen-Verse/DiscoveryRSI 🤗 Models: huggingface.co

Excited to introduce DiscoveryRSI! One model that recursively learns to discover valuable hypotheses, and gets sharper with every research round. Let's RSI Discovery! 📄 Paper: arxiv.org 💻 Code: Gen-Verse/DiscoveryRSI #RSI #AIforScience #AIAgents #ScientificDiscovery

Can research agents get better at science by learning from the research they do? Introducing DiscoveryRSI 🧬: a framework that turns hypothesis-driven research into a recursive self-improvement loop. 📄 Paper: arxiv.org 💻 Code: Gen-Verse/DiscoveryRSI 🔴 Live Arena: foundry-arena.rsi-arena.workers.dev 🔹 What is a valuable hypothesis? We define it as a joint object H = (claim, investigation plan, decision criteria, alternative explanations). A hypothesis is valuable if resolving it would change a research decision. It is testable if the plan can produce evidence that supports, weakens, or bounds the claim. 🔹 How it works - Lead, Generation, Evolution, Critique and Progress agents grow hypothesis trees through evolve / repair / crossover / restart - Every version stays linked to its evidence, data, assessments and contribution records - Selected plans compile into executable Harbor tasks 🔹 Three self-improvement mechanisms ① Hypothesis credit optimization: blinded TrueSkill assessments assign bounded credit to the agent steps that actually contributed, and DiscoveryRSI-27B is trained only on those steps (Qwen3.8-27B init, 256K context; 2,397 of 15,792 trajectories kept; 2.12% of tokens supervised) ② Harness replay: diagnose recurring failures such as repeated exploration, failed tool use, lost artifacts and premature closure, then refine the workflow ③ Rediscovery: run the accepted model and harness updates on fresh questions 🔹 Results (DFM-Bench, 4 life-science domains) - Hypothesis quality 4.86 → 5.18 → 5.39 over two rounds - Falsifiability 4.00 → 6.20, alternative explanations 4.44 → 5.98, data resolvability 3.73 → 4.99 - Evolve/repair share 77.5% → 91.7%, restarts 21.7% → 5.0%: the agent learns to keep developing an idea instead of starting over - Caveat: importance and novelty dip slightly. The model learns to design better tests, not yet to choose bolder questions. 🔹 Expert review 27 physicians and bioscience researchers scored 292 anonymized proposals: - 41% of DiscoveryRSI-27B revisions improve quality vs 31% (GPT-5.6 Sol) and 25% (GPT-6 Luna) - Only 12% make things worse vs 32% and 31% 🔹 Does it anticipate real science? Against 16,440 publications filtered by each model's cutoff, 6.3% of DiscoveryRSI-27B's hypotheses connect to later published research (16.9% in therapeutic targets). Comparison: DeepSeek 5.4%, GLM 2.8%, GPT 2.5%, Claude 2.3%. Watch six models do research in real time in DiscoveryRSI Arena, and follow every proposal, critique and compile step 👇 foundry-arena.rsi-arena.workers.dev Co-led by Jianyuan Zhong, Chenwei Wu & Zhaochen Yu, with wonderful collaborators across PhAI Labs, CUHK, Stanford, MIT, Harvard Medical School, Oxford, Princeton, PKU and more. Grateful to @james_y_zou @Jiaru_Zou @Charles_Y_Wu @zhenfei_yin_yzf 🙏

🚀 Introducing DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses 6.3% of DiscoveryRSI-27B's hypotheses match later published work, vs 2.5% GPT-6 & 2.3% Claude Opus 5. 📄 arxiv.org 💻 Gen-Verse/DiscoveryRSI 🔴 foundry-arena.rsi-arena.workers.dev

DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses Scientific discovery requires developing worthwhile hypotheses into practical investigations, yet learning from this evolving research process remains challenging. We introduce an operational definition of valuable and testable research hypotheses: evidence-grounded claims about consequential uncertainties, coupled with feasible investigations, explicit decision criteria, and alternative explanations. Our DiscoveryRSI framework develops hypothesis trees through multi-agent evolutionary search, compiles selected investigations into executable Harbor tasks, and recursively learns through three mechanisms: hypothesis credit optimization attributes scientific assessments to documented research contributions; harness replay identifies productive strategies and recurring failures to refine research workflows; and rediscovery applies accepted improvements to new investigations, generating further experience for subsequent learning. We also introduce DFM-Bench, which provides comprehensive assessment of hypothesis quality and cumulative research progress at the levels of hypotheses, revisions, development paths and research runs. Across four life-science domains, automated evaluation shows that hypothesis quality rises from 4.86 for the base model to 5.39 after two rounds of self-improvement, with the large gains in falsifiability and data resolvability. We further validate hypothesis quality through expert review, comparing our DiscoveryRSI-27B with other research agents. Preliminary human assessments show that DiscoveryRSI-27B scores higher than GPT-6 Luna on hypothesis importance and whether claims are plausible and evidence-bounded. DiscoveryRSI-27B achieved a 6.3% rate of hypothesis correspondence with research published after the model cutoff, largely outperform GPT-6 with 2.5% and Claude Opus 5 with 2.3%, while delivering greater human-rated improvement along research paths. Comments: Code: Gen-Verse/DiscoveryRSI