LLM4Science最新趋势 （吴郦军 微软研究院）.pdf

《LLM4Science最新趋势 （吴郦军 微软研究院）.pdf》由会员分享，可在线阅读，更多相关《LLM4Science最新趋势 （吴郦军 微软研究院）.pdf（21页珍藏版）》请在三个皮匠报告上搜索。

1、Recent Trends in LLM4ScienceLijun Wu*Works done when in MSR AI4Science,now in ByteDanceYSSNLP2024YSSNLP2024YSSNLP2024 YSSNLP2024YSSNLP2024YSSNLP2024Large Language ModelsLarge language models(LLMs)have shown great success!YSSNLP2024YSSNLP2024YSSNLP2024 YSSNLP2024YSSNLP2024YSSNLP2024Scientific Discove

2、ryAlphaFold(1,2,3)(protein folding)GNoMe(material discovery)GraphCast/MetNet/Aurora(weather)YSSNLP2024YSSNLP2024YSSNLP2024 YSSNLP2024YSSNLP2024YSSNLP2024Large language models(LLMs)Scientific discoveryLLMfor ScienceYSSNLP2024YSSNLP2024YSSNLP2024 YSSNLP2024YSSNLP2024YSSNLP2024Outline1.LLM4Science Repo

3、rt(GPT-4)2.Multimodal Pretrained Scientific LLM3.FutureYSSNLP2024YSSNLP2024YSSNLP2024 YSSNLP2024YSSNLP2024YSSNLP2024LLM4Science ReportThe Impact of Large Language Models on Scientific Discovery:a Preliminary Study using GPT-4MSR AI4Science,Microsoft Azure QuantumYSSNLP2024YSSNLP2024YSSNLP2024 YSSNLP

4、2024YSSNLP2024YSSNLP2024Understan-ding conceptsDrug-target bindingMolecular property predictionRetrosyn-thesis Novel molecule generationCoding assistanceUnderstan-ding bio-sequencesReasoning with bio-knowledgeDesigningBio-expPractical exampleMolecular dynamics simulationElectronic structureKnowing b

5、asic conceptsSolving PDEsAI for PDEsMemoriza-tion and designingCandidate proposalStructure generationProperty predictionSynthesis planningCoding assistanceDrug DiscoveryMaterials DesignComputational chemistryBiologyPDEAnalyzing literatureCode developmentData analysisTheoretical modelingMethod guidan

6、cePrediction Experimental designConcept clarificationHypothesis generationScientific AreasCapabilitiesThe Impact of Large Language Models on Scientific Discovery:a Preliminary Study using GPT-430+researchers,5 areas,9 capabilities,21 tasksYSSNLP2024YSSNLP2024YSSNLP2024 YSSNLP2024YSSNLP2024YSSNLP2024