Preeti
A semiautomated synthesis system was constructed, consisting of a dispensing system and rail-operated robotic arm. The system was built using affordable components from open-source 3D printers. Sample bottles with specialized containers and ArUco markers were used for chemical reactions. The system operated based on visual feedback, allowing for liquid handling, sample observation, and container movement. The accuracy of the dispensing device was verified experimentally.
The system was used to synthesize polyamic acid particles through precipitation polymerization. The system was not suitable for reactions requiring strict inert conditions or spectrophotometric devices. The particles obtained were unique and could be used for various applications.
GPT-4, a multimodal large language model, was used to automatically generate experimental notes and observations. The model could analyze images and provide textual observations of experimental notes. It could also estimate turbidity levels and particle sizes based on images captured during experiments.
Experiments varying reaction conditions showed differences in turbidity and particle size. Human-operated experiments yielded better results compared to robotic experiments due to factors like misalignment of sample bottles affecting stirring conditions.
The study highlighted the potential of foundational models in articulating nuanced situational judgments made in the laboratory setting. However, caution was advised in using language models for advanced considerations in chemical research. Future research efforts integrating information science with chemistry could enhance the capabilities of language models in chemical experiments.
Source link
Source link: https://www.nature.com/articles/s41428-024-00930-9
GIPHY App Key not set. Please check settings