OpenAI now offers the exciting opportunity to create personal GPTs and integrate them with proprietary data. This brings up crucial questions: How does OpenAI manage this data? More importantly, can users of GPTs access and download the proprietary data files they upload?
What exactly are GPTs?
GPTs, a new feature of ChatGPT, allow users to create custom ChatGPTs. Users can fine-tune their GPTs with custom instructions, upload additional data, and even connect their GPT to external software via the API.
This sounds like an extremely cool and useful feature, enhancing the value of ChatGPT. However, there is a caveat: you become a data provider to OpenAI.
Your Data or OpenAI's Data?
According to OpenAI's terms and conditions, you retain ownership of the data you input, along with the responses you receive. However, OpenAI may utilize your content to enhance, maintain, and develop their services, or to comply with legal requirements, unless you are eligible for and actually choose to opt out.
(see Terms of use (openai.com))
Currently, the definition of 'content' encompasses everything sent to OpenAI via ChatGPT, including uploaded files, inserted prompts, etc.
Opt-Out Confusion in GPTs
The opt-out feature in GPTs presents a challenge. While it permits creators to exclude conversation data from model training by deselecting an option in the 'Additional Settings' of GPTs configuration, it remains ambiguous whether this also applies to uploaded files.
Although the OpenAI support article (GPTs Data Privacy FAQs | OpenAI Help Center) suggests that opt-out could apply to both conversation data and uploaded files, the specifics of implementation remain unclear.
Can anyone access the files I upload to GPTs?
If you grant user access to your GPTs, it’s reasonable to assume they might access the files you upload. Essentially, this is part of sharing your knowledge, including data, with the world.
Downloading each uploaded file isn’t straightforward. However, there have been instances of savvy individuals accessing both files and other content. Additionally, some users of GPTs have reportedly accessed detailed descriptions from the Instructions.
Therefore, it's important to carefully consider when creating your GPTs: Are you fully comfortable sharing all uploaded content with your users? If not, you might need to reconsider building that GPT or review the uploaded content.
Who Can Access My GPTs?
To maintain some control over your data, you can decide who has access to your GPTs. ChatGPT offers three choices:
- Personal use only.
- Limited sharing via a link with select individuals.
- Public availability through the forthcoming GPTs marketplace.
If your uploaded data is sensitive, it's best not to share it. And if you plan to publish your GPTs globally, ensure that nothing confidential or private is included.
Conclusion
GPTs have simplified the process of enhancing ChatGPT with personal data and knowledge, significantly boosting its utility. However, this ease of use brings its own set of challenges, which can be easy to overlook.
As of now, OpenAI has not provided clear guidelines on how creators can opt out of model training. This ambiguity suggests that the data we upload could potentially be used for training future OpenAI models.
Moreover, users of your GPTs might gain direct access to your uploaded files and custom instructions, so caution is advised in this area.
Nevertheless, if your goal is to contribute positively to the world by sharing your knowledge with OpenAI and the users of your GPTs, GPTs have made this remarkably straightforward.