Dataset Card for ShareGPT52K90K

Dataset Description

• Homepage: N/A
• Repository: N/A
• Paper: N/A
• Leaderboard: N/A
• Point of Contact: Ronsor Labs ronsor@ronsor.com

Dataset Summary

This dataset is a collection of approximately 52,00090,000 conversations scraped via the ShareGPT API before it was shut down. These conversations include both user prompts and responses from OpenAI's ChatGPT.

This repository now contains the new 90K conversations version. The previous 52K may be found in the old/ directory.

Supported Tasks and Leaderboards

• text-generation

Languages

This dataset is expected to primarily consist of messages in English and other Western languages.

Dataset Structure

Data Instances

{
    "id": "2wtLdZm",
    "conversations": [
        {
            "from": "human",
            "value": "Are you familiar with the SAS programming language?"
        },
        {
            "from": "gpt",
            "value": "<div class=\"markdown prose w-full break-words dark:prose-invert light\"><p>Yes, I am familiar with the SAS programming language. SAS is a statistical software suite that is widely used in data analysis, data management, and business intelligence applications. It has its own programming language, which is used to manipulate and analyze data, generate reports, and automate data-driven workflows. The language is known for its ease of use and powerful capabilities, and it is commonly used in industries such as healthcare, finance, and government.</p></div>"
        },
        {
            "from": "human",
            "value": "Can you write a SAS proc that imports a dataset from a CSV saved on my local machine?"
        }
    ]
}

Data Fields

• id: conversation id
• conversations: conversation session array
  • from: ChatGPT ("gpt") or the user ("human")
  • value: message contents as raw HTML

Data Splits

N/A

Dataset Creation

Curation Rationale

This is a decently large dataset of realistic human-AI conversations which I believe should be released to the research community.

Source Data

Initial Data Collection and Normalization

This data was collected using the ShareGPT API.

Who are the source language producers?

ShareGPT users and OpenAI ChatGPT.

Annotations

Annotation process

N/A

Who are the annotators?

N/A

Personal and Sensitive Information

This dataset may contain personal information, if ShareGPT users were sending such information to ChatGPT. ChatGPT warns users not to submit personal information to it, however, so without further evaluation, we believe that this dataset should contain little or no personal information.

Considerations for Using the Data

Social Impact of Dataset

This dataset may be used to train models that are competitive with OpenAI's ChatGPT. Please filter this dataset first, as it may contain canned responses, raw HTML, and other undesirable information.

Discussion of Biases

This dataset exhibits all the biases of OpenAI's ChatGPT models (GPT-3.5 and GPT-4) as well as the biases of the users who uploaded the conversations.

Other Known Limitations

N/A

Additional Information

Dataset Curators

None.

Licensing Information

CC0: No Rights Reserved.

The output of machine learning algorithms is uncopyrightable in the United States and other jurisdictions. Additionally, the OpenAI terms of service do not apply to this dataset as users of this dataset are not accessing the OpenAI service.

Citation Information

TODO

Contributions

These conversations were allegedly scraped by an anonymous user on 4chan.

The 90K version was sourced from this post. Thanks, anon!