Reincarnate转发了
?? How can we use LLMs to accelerate scientific discovery in materials science and chemistry? Let's find out! This year, hundreds of people from across the globe worked together in a hackathon to BUILD groundbreaking prototypes — showing the path to breakthroughs in next generation batteries, sustainability, advanced computing, and more. The teams built 34 incredible prototypes in only 1 day, and showed potential for impacts across areas of: - Extracting and Organizing Knowledge - Improving LLM Property Prediction - Creation of Novel Human/Computer Interfaces - Automating tasks and Improving Efficiency Automation - Reducing Information Friction - Empowering Learners - Evaluating LLM Capabilities with Benchmarks ?? A few highlights ?? ?? Evaluating the capabilities of LLMs in materials science and helping students learn key concepts is challenging. MaSTeA developed an interactive web app for materials science question answering, and a way to generate an automated benchmark dataset to evaluate LLM capabilities to identify strengths and weaknesses in various subfields, enhancing educational and LLM tools.? Link: https://lnkd.in/geHWJUuM ?? LLMs work well with some forms of chemical information, e.g., SMILES strings, but may miss out on key molecular 3D information. Geometric Geniuses showed a project that integrated chemical feature vectors with LLMs to combine structural information, 3D geometric information, and LLMs for enhanced property prediction. Link: https://lnkd.in/gUrZYSGC ?? Computational chemistry is hard, and there are many tools that you need to learn to proficiently simulate materials. LangSim built a tool that provides a natural language interface for simulating materials at the atomic level, reducing the time and effort needed to learn and run simulation codes at the level of an expert. @RadicalAI first prize winner Link: https://lnkd.in/gEVUe3KE It’s important that researchers have shared term definitions, and GlossaGen shows the pathway. Glossagen built a tool that enables automated creation of glossaries for knowledge graphs from research papers. @RadicalAI_inc 2nd prize winner Link: ?https://lnkd.in/g7GTSNMW ?? Designing high performing MOFs will lead to better materials for catalysis, photovoltaics, sensing, drug delivery, energy storage, and more. PoreVoyant built an AI agent to collect guidelines for designing low band gap MOF materials from literature. The agent then continuously refines and explores the design space to determine the best candidates. @RadicalAI_inc third prize winner https://lnkd.in/gE3KekUC Read about all team projects here: https://lnkd.in/gFZDbkWn Code links here: https://lnkd.in/gfuPQ8_V