What Should Data Science Education Do with Large Language Models?

Yong Zheng

31 Juli 2023

ChatGPT, an implementation and application of large language models, has gained significant popularity since its initial release. Researchers have been exploring ways to harness the practical benefits of ChatGPT in real-world scenarios. Educational researchers have investigated its potential in various subjects, e.g., programming, mathematics, finance, clinical decision support, etc. However, there has been limited attention given to its application in data science education. This paper aims to bridge that gap by utilizing ChatGPT in a data science course, gathering perspectives from students, and presenting our experiences and feedback on using ChatGPT for teaching and learning in data science education. The findings not only distinguish data science education from other disciplines but also uncover new opportunities and challenges associated with incorporating ChatGPT into the data science curriculum.

Adalbert Gerald Soosai Raj Xinyi Ai Meenakshi Syamkumar + 2 lainnya

7 Maret 2024

ChatGPT is a conversational AI platform that can produce code to solve problems when provided with a natural language prompt. Prior work on similar AI models has shown that they perform well on typical intro-level Computer Science problems. However, little is known about the performance of such tools on Data Science (DS) problems. In this work, we assess the performance of ChatGPT on assignments from three DS courses with varying difficulty levels. First, we apply the raw assignment prompts provided to the students and find that ChatGPT performs well on assignments with dataset(s) descriptions and progressive question prompts, which divide the programming requirements into sub-problems. Then, we perform prompt engineering on the assignments for which ChatGPT had low performance. We find that the following prompt engineering techniques significantly increased ChatGPT's performance: breaking down abstract questions into steps, breaking down steps into multiple prompts, providing descriptions of the dataset(s), including algorithmic details, adding specific instructions to entice specific actions, and removing extraneous information. Finally, we discuss how our findings suggest potential changes to curriculum design of DS courses.

P. Ridge Paul Denny Stephen R. Piccolo + 2 lainnya

7 Maret 2023

Computer programming is a fundamental tool for life scientists, allowing them to carry out essential research tasks. However, despite various educational efforts, learning to write code can be a challenging endeavor for students and researchers in life-sciences disciplines. Recent advances in artificial intelligence have made it possible to translate human-language prompts to functional code, raising questions about whether these technologies can aid (or replace) life scientists’ efforts to write code. Using 184 programming exercises from an introductory-bioinformatics course, we evaluated the extent to which one such tool—OpenAI’s ChatGPT—could successfully complete programming tasks. ChatGPT solved 139 (75.5%) of the exercises on its first attempt. For the remaining exercises, we provided natural-language feedback to the model, prompting it to try different approaches. Within 7 or fewer attempts, ChatGPT solved 179 (97.3%) of the exercises. These findings have implications for life-sciences education and research. Instructors may need to adapt their pedagogical approaches and assessment techniques to account for these new capabilities that are available to the general public. For some programming tasks, researchers may be able to work in collaboration with machine-learning models to produce functional code.

Chenglin Wu Danyang Li Wei Tao + 22 lainnya

28 Februari 2024

Large Language Model (LLM)-based agents have shown effectiveness across many applications. However, their use in data science scenarios requiring solving long-term interconnected tasks, dynamic data adjustments and domain expertise remains challenging. Previous approaches primarily focus on individual tasks, making it difficult to assess the complete data science workflow. Moreover, they struggle to handle real-time changes in intermediate data and fail to adapt dynamically to evolving task dependencies inherent to data science problems. In this paper, we present Data Interpreter, an LLM-based agent designed to automatically solve various data science problems end-to-end. Our Data Interpreter incorporates two key modules: 1) Hierarchical Graph Modeling, which breaks down complex problems into manageable subproblems, enabling dynamic node generation and graph optimization; and 2) Programmable Node Generation, a technique that refines and verifies each subproblem to iteratively improve code generation results and robustness. Extensive experiments consistently demonstrate the superiority of Data Interpreter. On InfiAgent-DABench, it achieves a 25% performance boost, raising accuracy from 75.9% to 94.9%. For machine learning and open-ended tasks, it improves performance from 88% to 95%, and from 60% to 97%, respectively. Moreover, on the MATH dataset, Data Interpreter achieves remarkable performance with a 26% improvement compared to state-of-the-art baselines. The code is available at https://github.com/geekan/MetaGPT.

E. Silva Hossein Hassani

27 Maret 2023

ChatGPT, a conversational AI interface that utilizes natural language processing and machine learning algorithms, is taking the world by storm and is the buzzword across many sectors today. Given the likely impact of this model on data science, through this perspective article, we seek to provide an overview of the potential opportunities and challenges associated with using ChatGPT in data science, provide readers with a snapshot of its advantages, and stimulate interest in its use for data science projects. The paper discusses how ChatGPT can assist data scientists in automating various aspects of their workflow, including data cleaning and preprocessing, model training, and result interpretation. It also highlights how ChatGPT has the potential to provide new insights and improve decision-making processes by analyzing unstructured data. We then examine the advantages of ChatGPT’s architecture, including its ability to be fine-tuned for a wide range of language-related tasks and generate synthetic data. Limitations and issues are also addressed, particularly around concerns about bias and plagiarism when using ChatGPT. Overall, the paper concludes that the benefits outweigh the costs and ChatGPT has the potential to greatly enhance the productivity and accuracy of data science workflows and is likely to become an increasingly important tool for intelligence augmentation in the field of data science. ChatGPT can assist with a wide range of natural language processing tasks in data science, including language translation, sentiment analysis, and text classification. However, while ChatGPT can save time and resources compared to training a model from scratch, and can be fine-tuned for specific use cases, it may not perform well on certain tasks if it has not been specifically trained for them. Additionally, the output of ChatGPT may be difficult to interpret, which could pose challenges for decision-making in data science applications.