Our new paper has just been published in Digital Discovery: “Can We Automate Scientific Reasoning in Closed Loop Experiments using Large Language Models?” https://pubs.rsc.org/en/Content/ArticleLanding/2026/DD/D5DD00520E
In this study, we show that hybrids of Bayesian optimisation (BO) and large language models (LLMs) can navigate high-dimensional chemical space. In some cases, LLM only optimisers outperform BO or BO/LLM hybrids.
Prof. Cooper said: “I think this work has interesting ramifications. It’s one of the longer papers that I’ve ever published, in part because of the various control studies we did to dig into the limitations of these LLM-based methods. It’s also the first paper that I’ve coauthored with my son, Max, who did an internship in our group last summer.”
The paper can be accessed here: https://pubs.rsc.org/en/Content/ArticleLanding/2026/DD/D5DD00520E
An independent commentary on this work in archived form can be found here: