arXiv:2606.10587v1 Announce Type: new 
Abstract: Large language models (LLMs) are on the rise for accelerating scientific discovery, most recently in advanced tasks such as generating valid scientific hypotheses. Yet in many discovery settings, the goal is not to identify a single best hypothesis since validation can be noisy and expensive, and scientists benefit from a set of high-quality alternative hypotheses that hedge against downstream uncertainty for the best solutions. Nevertheless, commonly used evolutionary search recipes tend to prioritize optimization over exploration in hypothesis generation, and the resulting selection pressure during the search process leads to diversity collapse. Motivated by these limitations, we formulate hypothesis search as a sampling problem, where the objective is to efficiently produce diverse, high-quality hypotheses under a fixed validation budget. Building on this perspective, we propose \ours, an evolutionary framework inspired by the classical parallel tempering algorithm that searches hypotheses at multiple temperature levels and enables principled information exchange across temperatures to improve exploration without disrupting convergence. Across domains including molecular discovery, equation discovery, and algorithm discovery, our approach consistently improves both hypothesis quality and diversity under the same validation budget, and produces candidates that remain robust under more expensive downstream computational validations.

أدت التطورات الأخيرة في نماذج اللغة الكبيرة (LLMs) إلى تطوير إطار جديد يهدف إلى تحسين تنوع توليد الفرضيات العلمية. تتناول هذه الطريقة القيود المفروضة على الأساليب التقليدية التي غالبًا ما تعطي الأولوية للتحسين على الاستكشاف، مما يؤدي إلى نقص في الفرضيات المتنوعة. يسعى الإطار المقترح إلى إنتاج مجموعة متنوعة من الفرضيات عالية الجودة بكفاءة ضمن ميزانية تحقق ثابتة.

Los recientes avances en modelos de lenguaje de gran tamaño (LLMs) han llevado al desarrollo de un nuevo marco destinado a mejorar la diversidad en la generación de hipótesis científicas. Este enfoque aborda las limitaciones de los métodos tradicionales que a menudo priorizan la optimización sobre la exploración, lo que resulta en una falta de hipótesis diversas. El marco propuesto busca producir de manera eficiente una variedad de hipótesis de alta calidad dentro de un presupuesto de validación fijo.

Des avancées récentes dans les modèles de langage de grande taille (LLMs) ont conduit au développement d'un nouveau cadre visant à améliorer la diversité de la génération d'hypothèses scientifiques. Cette approche répond aux limitations des méthodes traditionnelles qui privilégient souvent l'optimisation au détriment de l'exploration, entraînant un manque d'hypothèses diversifiées. Le cadre proposé cherche à produire efficacement une variété d'hypothèses de haute qualité dans un budget de validation fixe.

Recent advancements in large language models (LLMs) have led to the development of a new framework aimed at enhancing the diversity of scientific hypothesis generation. This approach addresses the limitations of traditional methods that often prioritize optimization over exploration, resulting in a lack of diverse hypotheses. The proposed framework seeks to efficiently produce a variety of high-quality hypotheses within a fixed validation budget.