According to a new preprint on arXiv, scientific discovery can be understood as an optimization problem. The authors define a vast state space containing possible theories and experiments, and an evaluation criterion that weighs quality, novelty, and validity. This framing treats the search for knowledge as a meta-optimization task, where the goal is to navigate that space efficiently.

The paper, titled 'Scientific discovery as meta-optimization: a combinatorial optimization case study,' applies this perspective to a concrete instance of combinatorial optimization. Large language models appear to play a role in the proposed approach, though the abstract does not specify how they are used or what results were obtained. The work is announced as a replace-cross update, suggesting it has been revised.

As the abstract is short, the details of the method and findings are not available from this summary alone. The key contribution, based on the abstract, is the explicit framing of discovery as an optimization problem with defined criteria.