{"slug": "training-reasoning-models-to-generate-diverse-scientific-ideas", "title": "Training Reasoning Models to Generate Diverse Scientific Ideas", "summary": "A paper submitted to arXiv on 15 October 2026 reports that an 8B-parameter model trained with a set-level reinforcement learning objective achieved higher creative utility — a measure combining quality and diversity — than Claude Opus 4.8 and GPT-6 Astra on the authors' contamination-free benchmark for document-grounded scientific ideation. In the training setup, the model writes K structured idea cards grounded in source papers in a single generation, and each idea's quality is weighted by its distance from the ideas written before it, steering new ideas away from predecessors. In a pilot human study with four experts, the trained model's idea sets were judged as promising as Claude Opus 4.8's and contained marginally more unique ideas, and when a coding agent ran proof-of-concept experiments, the model's success@k was comparable to Claude Opus 4.8's, indicating the added diversity did not cost executability.", "body_md": "# Computer Science > Computation and Language\n\n  [Submitted on 15 Oct 2026]\n\n# Title:Training Reasoning Models to Generate Diverse Scientific Ideas\n\n[View PDF](https://arxiv.org/pdf/2610.22324)\n\n[HTML (experimental)](https://arxiv.org/html/2610.22324v1)\n\nAbstract:One of the first steps of a research pipeline is generating ideas. Large Language Models (LLMs) are increasingly used to automate this step, but most of this work optimizes the quality of individual ideas, while overlooking diversity. In our work, we study document-grounded ideation, focusing on increasing the diversity of LLM generated ideas and ensuring they are high quality. Given prior papers, a model must propose a set of ideas that are each high-quality and clearly distinct from one another, while also being grounded in the input documents. To optimize for this, we propose training with a set-level reinforcement learning objective. In our setup, in a single generation, the model writes K structured idea cards grounded in the source papers, and the set is scored jointly, where each idea's quality is weighted by its distance from the ideas written before it. This teaches the model to steer each new idea away from its predecessors. We show that our trained 8B model achieves higher creative utility (a measure combining quality and diversity) than Claude Opus 4.8 and GPT-6 Astra on our contamination-free benchmark. In a pilot human study with four experts, our sets are judged as promising as Claude Opus 4.8's and contain marginally more unique ideas. When a coding agent runs proof-of-concept experiments on the ideas, our model's success@k is comparable to Claude Opus 4.8's, so we see that the added diversity does not cost any executability.\n    \n\n### Additional Features\n\n### References & Citations\n\nLoading...\n\n# Bibliographic and Citation Tools\n\nBibliographic Explorer \n\n*(*[What is the Explorer?](https://info.arxiv.org/labs/showcase.html#arxiv-bibliographic-explorer))\nConnected Papers \n\n*(*[What is Connected Papers?](https://www.connectedpapers.com/about))\nLitmaps \n\n*(*[What is Litmaps?](https://www.litmaps.co/))\nscite Smart Citations \n\n*(*[What are Smart Citations?](https://www.scite.ai/))\n# Code, Data and Media Associated with this Article\n\nalphaXiv \n\n*(*[What is alphaXiv?](https://alphaxiv.org/))\nCatalyzeX Code Finder for Papers \n\n*(*[What is CatalyzeX?](https://www.catalyzex.com))\nDagsHub \n\n*(*[What is DagsHub?](https://dagshub.com/))\nGotit.pub \n\n*(*[What is GotitPub?](http://gotit.pub/faq))\nHugging Face \n\n*(*[What is Huggingface?](https://huggingface.co/huggingface))\nScienceCast \n\n*(*[What is ScienceCast?](https://sciencecast.org/welcome))\n# Demos\n\n# Recommenders and Search Tools\n\nInfluence Flower \n\n*(*[What are Influence Flowers?](https://influencemap.cmlab.dev/))\nCORE Recommender \n\n*(*[What is CORE?](https://core.ac.uk/services/recommender))\n# arXivLabs: experimental projects with community collaborators\n\narXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.\n\nBoth individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.\n\nHave an idea for a project that will add value for arXiv's community? [**Learn more about arXivLabs**](https://info.arxiv.org/labs/index.html).", "url": "https://wpnews.pro/news/training-reasoning-models-to-generate-diverse-scientific-ideas", "canonical_source": "https://arxiv.org/abs/2610.22324", "published_at": "2026-10-18 10:50:27+00:00", "updated_at": "2026-10-18 11:48:29.067551+00:00", "lang": "en", "topics": ["ai-research", "large-language-models", "machine-learning", "artificial-intelligence"], "entities": ["arXiv", "Claude Opus 4.8", "GPT-6 Astra", "Anthropic", "OpenAI"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/training-reasoning-models-to-generate-diverse-scientific-ideas", "markdown": "https://wpnews.pro/news/training-reasoning-models-to-generate-diverse-scientific-ideas.md", "text": "https://wpnews.pro/news/training-reasoning-models-to-generate-diverse-scientific-ideas.txt", "jsonld": "https://wpnews.pro/news/training-reasoning-models-to-generate-diverse-scientific-ideas.jsonld"}}