188宝金博页面版

ChatGPT and Environmental ResearchJun-Jie Zhu, Jinyue Jiang, Meiqi Yang, and Zhiyong Jason Ren*Cite This: https://doi.org/10.1021/acs.est.3c01818 Read OnlineACCESSMetrics & More Article Recommendations * s? Supporting InformationKEYWORDS: ChatGPT, benef i ts, opportunities, over-reliance■CHATGPT: A CONVERSATIONAL AI THAT WILLFUNDAMENTALLY CHANGE THE WAY WE FINDINFORMATIONChatGPT, the latest text-based artif i cial intelligence (AI) tool,has quickly gained popularity and is poised to revolutionizevarious aspects of our lives, including education and research.With its advanced natural language processing (NLP)capabilities, ChatGPT can understand and interpret humanlanguage like never before, allowing users to ask questions andreceive answers in a conversational and intuitive manner. Inthis Viewpoint, we aim to draw from our NLP researchbackground and share our experience and thoughts aboutChatGPT by providing 10 real-world examples from dif f erentareas of environmental research. Our objective is todemonstrate how this emerging tool can be leveraged forresearch purposes while also highlighting potential pitfalls andchallenges. By sharing these experiences, we hope to encouragethe responsible and ef f ective use of ChatGPT in research andbeyond.The generative pretrained transformer (GPT) is a cutting-edge natural language generation (NLG) model, and its latestiteration, GPT-3.5 (GPT-4 1 was released on March 14, 2023),was on a massive corpus of textual data, such as books, articles,and Web sites, with billions of model parameters (GPT-3 forthe details). 2 ChatGPT is a f i ne-tuned application based on theGPT-3.5 engine at its initial release that uses supervised f i ne-tuning modeling (learning based on labeled prompt data),reward model construction (ranking the model responses), andproximal policy optimization (a class of reinforcement learningto optimize the reward policy). Two techniques used inChatGPT are in-context learning and prompt engineering. In-context learning enables the agent to learn and adapt in realtime, making it more versatile and capable of handling a widerrange of situations. While ChatGPT can respond to a questionwith no additional hints (zero-shot prompts), its responsequality improves by providing additional examples beforeasking questions (few-shot prompts). Prompt engineeringinvolves designing model inputs, such as questions andstatements, to obtain better outputs (i.e., responses).The popularity of ChatGPT stems from its rapid,informative, and seemingly “intelligent” responses to anyquestions. However, it is important to question whether themodel truly understands the content it produces, becausemistakes and errors are also frequently obtained even forsimple questions. As a result, it is essential to exercise cautionand avoid over- and underestimating the potential andcapability of the emerging tool (Figure 1).■WHAT WE FOUND BENEFICIAL AND WHERE WESEE POTENTIALWriting Improvement, Key Points, and ThemeIdentif i cation. Writing is an essential part of research, andwhile ChatGPT should not be relied upon as the originalcontent provider, it does provide great benef i ts in languagepolishing and identifying errors. This capability is particularlyvaluable for non-native speakers of the intended language.Moreover, ChatGPT can be used to translate text with morecustomized requests, and it can be utilized to summarizecritical information from lengthy material. For example, byasking ChatGPT to identify key points, synopses, and mainthemes of a lengthy text, researchers can save time and gain agood understanding of all of the material to facilitate bettercomprehension (example S1). Alternatively, ChatGPT canSpecial Issue: Data Science for Advancing Environ-mental Science, Engineering, and TechnologyPublished: March 21, 2023Figure 1. We should leverage the advantages (right side) thatChatGPT of f ers us while also exercising caution regarding potentialhidden pitfalls (left side).Viewpoint pubs.acs.org/est© 2023 The Authors. Published byAmerican Chemical SocietyAhttps://doi.org/10.1021/acs.est.3c01818Environ. Sci. Technol. XXXX, XXX, XXX−XXXDownloaded via CHINA ACAD ENGINEERING PHYSICS on March 24, 2023 at 04:48:22 (UTC).See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.