Next month, Microsoft’s GitHub will commence a new initiative where customer interaction data, including code snippets and associated context, will be harnessed to enhance its AI models.
This policy change affects users of Copilot Free, Pro, and Pro+, starting April 24. However, those utilizing Copilot Business or Copilot Enterprise will not be subject to this change. Additionally, students and educators using Copilot will remain unaffected.
For those whose data will be used, opting out is possible by visiting /settings/copilot/features and deselecting the option that allows GitHub to utilize their data for training AI models.
GitHub’s Chief Product Officer has expressed a preference that users remain opted in.
He emphasized that participating would enhance the AI’s understanding of development procedures, improving the accuracy of code suggestions and aiding in bug detection prior to deployment.
GitHub’s FAQs point out that this approach is not unique; several companies have similar policies regarding data use.
Rodriguez argues that using interaction data can significantly enhance AI models, citing improvements observed when integrating data from Microsoft staff.
The specific types of data GitHub intends to collect include:
- Accepted or modified model outputs;
- Displayed code snippets;
- Contextual information surrounding the cursor;
- User-generated comments and documentation;
- File naming and repository structure;
- Engagement with Copilot features;
- User feedback, such as ratings.
This shift in policy could change perceptions around GitHub private repositories. Even in private settings, data may still be utilized for AI training while the user interacts with Copilot.
If a user’s settings permit it, code snippets from private repositories could be collected during the Copilot sessions, which has sparked skepticism among the GitHub community.
Current discussions around this policy change have seen minimal support. Among user-generated comments, only one appears to have publicly endorsed the proposal.
Some might temper their concerns upon realizing that OpenAI’s Codex—integral to GitHub Copilot—has already been trained on a plethora of publicly accessible code.
Halting such data collection seems unlikely to alter the fundamental dynamics of an AI-centric marketplace built on widespread data utilization, often without explicit consent.
Key Takeaways
- GitHub is introducing a new policy to use customer interaction data for AI training.
- Users can opt out of data collection, but the consent model is U.S.-centric.
- Increased AI capabilities are promised through the use of this data.
- Private data usage may blur the traditional boundaries of repository confidentiality.
- User feedback has largely been negative, reflecting concern and mistrust.
- Understanding previous data practices in AI could alleviate some apprehensions.