OpenAI is set to initiate the testing phase for a new feature within ChatGPT that enables the chatbot to interface with various platforms. This new feature, referred to as ChatGPT Connectors, will be accessible to users subscribed to the ChatGPT Teams service. Initially, the tool is expected to support integrations with Google Drive for Workspace and the communications platform Slack, allowing ChatGPT to leverage internal data to respond to user inquiries based on that information.
OpenAI to Test ChatGPT Connectors
A report from TechCrunch reveals that OpenAI plans to begin testing this feature shortly for its Teams subscribers. Designed as an enterprise tool, ChatGPT Connectors aims to link with third-party databases and communication tools, enabling users to retrieve information and address queries originating from those sources.
According to an internal document referenced by the publication, beta testing for the ChatGPT Connectors feature is imminent. Initially, it will function with Google Drive and Slack, with potential for future integration with Microsoft’s SharePoint and Box. This feature, exclusive to ChatGPT Teams, will capably extract text data from a range of sources, including files, presentations, spreadsheets, and conversations within Slack.
Powered by a variant of the GPT-4o AI model, ChatGPT Connectors aims to be grounded in the “internal knowledge” of the connected platforms. All members within a participating ChatGPT Team workspace are expected to access the model through OpenAI’s chatbot.
However, there are concerns regarding privacy with this new feature. The report highlights that the custom GPT-4o AI model navigates through internal databases to create a search index by syncing encrypted versions of files and conversations with OpenAI’s servers. Details regarding the duration of file storage and potential access by third parties remain unclear.
Moreover, ChatGPT Connectors is designed to provide sourcing for information that isn’t directly referenced in the answers supplied. A button indicating this additional information will be located at the bottom of each response. The AI model will have the capability to access both external online information and the data from its training set.
The internal documentation emphasizes that the new feature will fully honor permissions set within Slack and Google Drive. It will specifically handle text inputs, while other formats such as images, audio, or video files will not be supported. Additionally, it will not have access to Slack’s direct messages or group chats, being limited to channel interactions.
OpenAI is reportedly inviting companies interested in participating in the beta test to submit 100 documents and prior Slack channel conversations. The firm asserts that this data will not be used to train the AI model directly, but it may inform synthetic data generation that could contribute to future training initiatives.
