Dataset Card for chinese_chatgpt_corpus

  Dataset Summary

This repo collects chinese corpus for Supervised Finetuning (SFT) and Reinforcement Learning From Human Feedback (RLHF).

  Supported Tasks and Leaderboards

More Information Needed

  Languages

Chinese

  Dataset Structure





  Data Instances





  train_data_external_v1.jsonl

Size of downloaded dataset files: 5.04 GB Size of the generated dataset: 0 GB Total amount of disk used:… See the full description on the dataset page: https://huggingface.co/datasets/sunzeyeah/chinese_chatgpt_corpus.

The dataset was compiled to examine the use of ChatGPT 3.5 in educational settings, particularly for creating and personalizing concept maps. The data has been organized into three folders: Maps, Texts, and Questionnaires. The Maps folder contains the graphical representation of the concept maps and the PlanUML code for drawing them in Italian and English. The Texts folder contains the source text used as input for the map's creation The Questionnaires folder includes the students' responses to the three administered questionnaires.

The researcher tests the QA capability of ChatGPT in the medical field from the following aspects:1. Test their reserve capacity for medical knowledge2. Check their ability to read literature and understand medical literature3. Test their ability of auxiliary diagnosis after reading case data4. Test its error correction ability for case data5. Test its ability to standardize medical terms6. Test their evaluation ability to experts7. Check their ability to evaluate medical institutionsThe conclusion is:ChatGPT has great potential in the application of medical and health care, and may directly replace human beings or even professionals at a certain level in some fields;The researcher preliminarily believe that ChatGPT has basic medical knowledge and the ability of multiple rounds of dialogue, and its ability to understand Chinese is not weak;ChatGPT has the ability to read, understand and correct cases;ChatGPT has the ability of information extraction and terminology standardization, and is quite excellent;ChatGPT has the reasoning ability of medical knowledge;ChatGPT has the ability of continuous learning. After continuous training, its level has improved significantly;ChatGPT does not have the academic evaluation ability of Chinese medical talents, and the results are not ideal;ChatGPT does not have the academic evaluation ability of Chinese medical institutions, and the results are not ideal;ChatGPT is an epoch-making product, which can become a useful assistant for medical diagnosis and treatment, knowledge service, literature reading, review and paper writing.

IntroductionIn recent years, numerous AI tools have been employed to equip learners with diverse technical skills such as coding, data analysis, and other competencies related to computational sciences. However, the desired outcomes have not been consistently achieved. This study aims to analyze the perspectives of students and professionals from non-computational fields on the use of generative AI tools, augmented with visualization support, to tackle data analytics projects. The focus is on promoting the development of coding skills and fostering a deep understanding of the solutions generated. Consequently, our research seeks to introduce innovative approaches for incorporating visualization and generative AI tools into educational practices.MethodsThis article examines how learners perform and their perspectives when using traditional tools vs. LLM-based tools to acquire data analytics skills. To explore this, we conducted a case study with a cohort of 59 participants among students and professionals without computational thinking skills. These participants developed a data analytics project in the context of a Data Analytics short session. Our case study focused on examining the participants' performance using traditional programming tools, ChatGPT, and LIDA with GPT as an advanced generative AI tool.ResultsThe results shown the transformative potential of approaches based on integrating advanced generative AI tools like GPT with specialized frameworks such as LIDA. The higher levels of participant preference indicate the superiority of these approaches over traditional development methods. Additionally, our findings suggest that the learning curves for the different approaches vary significantly. Since learners encountered technical difficulties in developing the project and interpreting the results. Our findings suggest that the integration of LIDA with GPT can significantly enhance the learning of advanced skills, especially those related to data analytics. We aim to establish this study as a foundation for the methodical adoption of generative AI tools in educational settings, paving the way for more effective and comprehensive training in these critical areas.DiscussionIt is important to highlight that when using general-purpose generative AI tools such as ChatGPT, users must be aware of the data analytics process and take responsibility for filtering out potential errors or incompleteness in the requirements of a data analytics project. These deficiencies can be mitigated by using more advanced tools specialized in supporting data analytics tasks, such as LIDA with GPT. However, users still need advanced programming knowledge to properly configure this connection via API. There is a significant opportunity for generative AI tools to improve their performance, providing accurate, complete, and convincing results for data analytics projects, thereby increasing user confidence in adopting these technologies. We hope this work underscores the opportunities and needs for integrating advanced LLMs into educational practices, particularly in developing computational thinking skills.

along with the corresponding answers from students and ChatGPT.

We implement the calculation of cosine similarity using the sklearn package [45].

GPT-3's water consumption for the training phase was estimated at roughly 4.8 billion liters of water, when assuming the model was trained on Microsoft's Iowa data center (OpeanAI has disclosed that the data center was used for training parts of the GPT-4 model). If the model were to have been fully trained in the Washington data center, water consumption could have been as high as 15 billion liters. That would've amounted to more than Microsoft's total water withdrawals in 2023.

This is the raw data that is used in the publication: ChatGPT as an education and learning tool for engineering, technology and general studies: performance analysis of ChatGPT 3.0 on CSE, GATE and JEE examinations of India.

Abstract:This dataset presents survey responses from first-year engineering students on their use of ChatGPT and other AI tools in a project-based learning environment. Collected as part of a study on AI’s role in engineering education, the data captures key insights into how students utilize ChatGPT for coding assistance, conceptual understanding, and collaborative work. The dataset includes responses on frequency of AI usage, perceived benefits and challenges, ethical concerns, and the impact of AI on learning outcomes and problem-solving skills.With AI increasingly integrated into education, this dataset provides valuable empirical evidence for researchers, educators, and policymakers interested in AI-assisted learning, STEM education, and academic integrity. It enables further analysis of student perceptions, responsible AI use, and the evolving role of generative AI in higher education.By making this dataset publicly available, we aim to support future research on AI literacy, pedagogy, and best practices for integrating AI into engineering and science curricula..................................................................................................................................................................Related PublicationThis dataset supports the findings presented in the following peer-reviewed article:ChatGPT in Engineering Education: A Breakthrough or a Challenge?Davood KhodadadPublished: 7 May 2025 | Physics Education, Volume 60, Number 4© 2025 The Author(s). Published by IOP Publishing LtdCitation: Davood Khodadad 2025 Phys. Educ. 60 045006DOI: 10.1088/1361-6552/add073If you use or reference this dataset, please consider citing the above publication......................................................................................................................................................................Description of the data and file structureTitle: ChatGPT in Engineering Education: Survey Data on AI Usage, Learning Impact, and CollaborationDescription of Data Collection:This dataset was collected through a survey distributed via the Canvas learning platform following the completion of group projects in an introductory engineering course. The survey aimed to investigate how students engaged with ChatGPT and other AI tools in a project-based learning environment, particularly in relation to coding, report writing, idea generation, and collaboration.The survey consisted of 15 questions:12 multiple-choice questions to capture quantitative insights on AI usage patterns, frequency, and perceived benefits.3 open-ended questions to collect qualitative perspectives on challenges, ethical concerns, and students' reflections on AI-assisted learning.Key areas assessed in the survey include:Students’ prior familiarity with AI tools before the course.Frequency and purpose of ChatGPT usage (e.g., coding assistance, conceptual learning, collaboration).Perceived benefits and limitations of using AI tools in an engineering learning environment.Ethical considerations, including concerns about over-reliance and academic integrity.The dataset provides valuable empirical insights into the evolving role of AI in STEM education and can support further research on AI-assisted learning, responsible AI usage, and best practices for integrating AI tools in engineering education.

In recent years, artificial intelligence (AI) has been increasingly used in education and supports teachers in creating educational material and students in their learning progress. AI- driven learning support has recently been further strengthened by the release of ChatGPT, in which users can retrieve expla- nations for various concepts in a few minutes through chat. However, to what extent the use of AI models, such as ChatGPT, is suitable for the creation of didactically and content-wise good exercises for programming courses is not yet known. Therefore, in this paper, we investigate the use of AI-generated exercises for beginner and intermediate programming courses in higher education using ChatGPT. We created 12 exercise sheets with ChatGPT for a beginner to intermediate programming course focusing on the objects-first approach. We report our process, prompts, and experience using ChatGPT for this task and outline good practices we identified. The generated exercises are assessed and revised, primarily using ChatGPT, until they met the requirements of the programming course. We assessed the quality of these exercises by using them in our course as external teaching assignment at the University of Education Ludwigsburg and let the students evaluate them. Results indicate the quality of the generated exercises and the time-saving for creating them using ChatGPT. However, our experience showed that while it is fast to generate a good version of an exercise, almost every exercise requires minor manual changes to improve its quality.

Many NLP applications require manual text annotations for a variety of tasks, notably to train classifiers or evaluate the performance of unsupervised models. Depending on the size and degree of complexity, the tasks may be conducted by crowd-workers on platforms such as MTurk as well as trained annotators, such as research assistants. Using four samples of tweets and news articles (n = 6,183), we show that ChatGPT outperforms crowd-workers for several annotation tasks, including relevance, stance, topics, and frame detection. Across the four datasets, the zero-shot accuracy of ChatGPT exceeds that of crowd-workers by about 25 percentage points on average, while ChatGPT's intercoder agreement exceeds that of both crowd-workers and trained annotators for all tasks. Moreover, the per-annotation cost of ChatGPT is less than $0.003---about thirty times cheaper than MTurk. These results demonstrate the potential of large language models to drastically increase the efficiency of text classification.

This is version 2 of the dataset created and used to explore ChatGPT-3.5's ability to write, justify and analyse English poems. This version was created after the reviewers decision that this paper may be published, if some changes are made.

The purpose of the research was to determine if ChatGPT-3.5 would be adopted in English poetry classrooms. As none of the theoretical models were applicable, the Artificial Intelligence Adoption Prediction Model (AIAPM) was designed. Based on this model, an Artificial Intelligence Adoption Prediction tool (AIAPT) was designed to calculate an Adoption Prediction Score (APS). Then, ChatGPT-3.5's ability to write, justify and analyse poems were explored.

It was found that ChatGPT-3.5 could write, justify, and analyse poems, but it could also make errors and hallucinate convincingly. Thus, the AIAPT was used to calculate the Adoption Prediction Score. The APS was 9, thus all factors of the AIAPM could drive the adoption decision. Thus, it could be predicted that ChatGPT-3.5 would be adopted in English poetry classrooms, both for ethical and unethical purposes. Based on the results, a few pro-active strategies were suggested.

This dataset contains all data created and used during the research, including the poems which were integrated in the paper: "An Artificial Intelligence Adoption Prediction Model to determine if ChatGPT-3.5 would be adopted in English poetry classrooms" which was submitted toe Heliyon for publication.

The data set records the perceptions of Bangladeshi university students on the influence that AI tools, especially ChatGPT, have on their academic practices, learning experiences, and problem-solving abilities. The varying role of AI in education, which covers common usage statistics, what AI does to our creative abilities, its impact on our learning, and whether it could invade our privacy. This dataset reveals perspective on how AI tools are changing education in the country and offering valuable information for researchers, educators, policymakers, to understand trends, challenges, and opportunities in the adoption of AI in the academic contex.

Methodology Data Collection Method: Online survey using google from Participants: A total of 3,512 students from various Bangladeshi universities participated. Survey Questions:The survey included questions on demographic information, frequency of AI tool usage, perceived benefits, concerns regarding privacy, and impacts on creativity and learning.

Sampling Technique: Random sampling of university students Data Collection Period: June 2024 to December 2024

Privacy Compliance This dataset has been anonymized to remove any personally identifiable information (PII). It adheres to relevant privacy regulations to ensure the confidentiality of participants.

For further inquiries, please contact: Name: Md Jhirul Islam, Daffodil International University Email: jhirul15-4063@diu.edu.bd Phone: 01316317573

Large Language Model content safety considerations text data, about 570,000 in total, this dataset can be used for tasks such as LLM training, chatgpt

This file contains all conversations with ChatGPT regarding its ability to help improving systems thinking skills incl. causal loop diagrams and system dynamics modelling.

Data on ChatGPT 3's and ChatGPT 4's performance on self-assessment questions for dentistry (SFLEDM) and allergy and clinical immunology (EEAACI), sourced from the University of Bern’s Institute for Medical Education platform.

We tested ChatGPT on 25 tasks focusing on solving common NLP problems and requiring analytical reasoning. These tasks include (1) a relatively simple binary classification of texts like spam, humor, sarcasm, aggression detection, or grammatical correctness of the text; (2) a more complex multiclass and multi-label classification of texts such as sentiment analysis, emotion recognition; (3) reasoning with the personal context, i.e., personalized versions of the problems that make use of additional information about text perception of a given user (user’s examples provided to ChatGPT); (4) semantic annotation and acceptance of the text going towards natural language understanding (NLU) like word sense disambiguation (WSD), and (5) answering questions based on the input text. More information in the paper: https://www.sciencedirect.com/science/article/pii/S156625352300177X

Artificial intelligence (AI) is one of the most common and essential
technologies in this modern era, especially in the education and research
sectors. It mimics machine-processed human intellect. In modern times,
ChatGPT is one of the most effective and beneficial tools developed by
OpenAI. Provides prompt answers and feedback to help academics and
researchers. Using ChatGPT has various advantages, including improving
methods of instruction, preparing interactive lessons, assessment, and
advanced problem-solving. Threats against ChatGPT, however, include
diminishing creativity, and analytical thinking. Additionally, students would
adopt unfair procedures when submitting any tests or assignments online,
which would increase their dependency on AI systems rather than thinking
analytically. In this study, we have demonstrated arguments on both sides of
AI technology. We believe that our study would provide a depth of
knowledge and more informed discussion. Data is collected via an offline
platform and then machine learning algorithms such as K-nearest neighbour
(K-NN), support vector machine (SVM), naive bayes (NB), decision tree
(DT), and random forest (RF) are used to analyze the data which helps to
improve teaching and learning techniques where SVM shows best
performance. The results of the study would offer several significant
learning and research directions as well as ensure safe and responsible
adoption.

BackgroundClinical data is instrumental to medical research, machine learning (ML) model development, and advancing surgical care, but access is often constrained by privacy regulations and missing data. Synthetic data offers a promising solution to preserve privacy while enabling broader data access. Recent advances in large language models (LLMs) provide an opportunity to generate synthetic data with reduced reliance on domain expertise, computational resources, and pre-training.ObjectiveThis study aims to assess the feasibility of generating realistic tabular clinical data with OpenAI’s GPT-4o using zero-shot prompting, and evaluate the fidelity of LLM-generated data by comparing its statistical properties to the Vital Signs DataBase (VitalDB), a real-world open-source perioperative dataset.MethodsIn Phase 1, GPT-4o was prompted to generate a dataset with qualitative descriptions of 13 clinical parameters. The resultant data was assessed for general errors, plausibility of outputs, and cross-verification of related parameters. In Phase 2, GPT-4o was prompted to generate a dataset using descriptive statistics of the VitalDB dataset. Fidelity was assessed using two-sample t-tests, two-sample proportion tests, and 95% confidence interval (CI) overlap.ResultsIn Phase 1, GPT-4o generated a complete and structured dataset comprising 6,166 case files. The dataset was plausible in range and correctly calculated body mass index for all case files based on respective heights and weights. Statistical comparison between the LLM-generated datasets and VitalDB revealed that Phase 2 data achieved significant fidelity. Phase 2 data demonstrated statistical similarity in 12/13 (92.31%) parameters, whereby no statistically significant differences were observed in 6/6 (100.0%) categorical/binary and 6/7 (85.71%) continuous parameters. Overlap of 95% CIs were observed in 6/7 (85.71%) continuous parameters.ConclusionZero-shot prompting with GPT-4o can generate realistic tabular synthetic datasets, which can replicate key statistical properties of real-world perioperative data. This study highlights the potential of LLMs as a novel and accessible modality for synthetic data generation, which may address critical barriers in clinical data access and eliminate the need for technical expertise, extensive computational resources, and pre-training. Further research is warranted to enhance fidelity and investigate the use of LLMs to amplify and augment datasets, preserve multivariate relationships, and train robust ML models.

Chang, C.-Y., Chen, I.-H., & Tang, K.-Y. (2024, in press). Roles and research trends of ChatGPT-based learning: A bibliometric analysis and systematic review. Educational Technology & Society. Data(n=50)