Track synthesis, materials and catalysis as it is published.
Search papers, labs, and topics across Lattice.
To overcome the low-entropy bias that causes LLMs to produce homogeneous scientific ideas, the authors develop AI Night-Scientist, a framework that trains models via Group Relative Policy Optimization (GRPO) to execute unstructured, serendipitous "night science" exploration. Grounded in cognitive science, the approach models creativity along action, process, and outcome axes to provide targeted semantic guidance rather than relying on stochastic decoding heuristics. The resulting policy expands scientific research direction diversity by 27.8% and contribution types by 14.9%, while improving proposal originality by 66.2 percentage points and predicted citation impact by up to 32.0 points.
Cracking LLM ideation requires more than high sampling temperatures: reinforcement learning over cognitive creativity axes expands scientific exploration diversity by 28% while boosting proposal originality by 66 percentage points.
Large language models (LLMs) excel at structured, verifiable tasks, but their low-entropy bias can produce homogeneous and predictable outputs, limiting their utility for open-ended scientific ideation. Effective discovery, however, spans a broader creative spectrum: from structured day science to loosely structured, serendipitous night science that reaches ideas beyond those typically considered. We introduce AI Night-Scientist, an agentic framework that uses reinforcement learning to teach models when and how to depart from predictable reasoning. Grounded in cognitive science, we model creativity along three axes: action (what to do and how creatively), process (when to explore versus exploit), and outcome (the novelty and usefulness of the resulting idea). We use these axes to train models with GRPO, exposing them to varying degrees and forms of creativity throughout training. This produces substantially more diverse scientific proposals, expanding the range of research directions by 27.8% and contribution types by 14.9% over the base model. It also improves predicted citation impact by up to 32.0 percentage points and originality by 66.2 points. These gains cannot be reproduced by simply increasing decoding temperature; instead, we find that semantic guidance specifying what kind of creativity to pursue is critical. Overall, our results suggest that creativity is a learnable, multi-level ability that can be shaped to help researchers reach ideas beyond those typically explored by LLMs.
Microsoft Research