Emiel van Miltenburg

One worry I sometimes have about chatgpt and other models where you only have indirect access to the model: what happens with your prompts?

If the prompts are stored, how can/will they be used? Are users just doing the dirty work for OpenAI et al.?

I mean, if one were to annotate all prompts from all users, it might be possible to crowdsource appropriate responses & train the model on that. This makes it harder to assess the "true" capacity of future models.

Jan 2, 2023 at 9:17pmWeb
Show threadMark DingemanseJan 2, 2023

Oh yes. I've seen folks here hard at work for OpenAI trying prompt variations and dutifully hitting 👍 and 👎

The #chatgpt FAQ couldn't be clearer: all prompts and 'conversations' are harvested for training, cannot be retrieved by you, and can only be deleted in bulk by relinquishing access (a dark pattern clearly designed to keep you on board)

PSA: "free research access" to ChatGPT by OpenAI means fuelling *their* research and giving *them* free access to distributed human cognition

Show threadEmiel van MiltenburgJan 2, 2023
@dingemansemark thanks for this, I hadn't seen this before. Glad I didn't use it yet! So many ways this data could be (ab)used..
Show threadSimone RescioJan 2, 2023
@dingemansemark yet another reminder that if a complex service is “free to use” you are the product
Show threadSimon WillisonJan 2, 2023
@dingemansemark it looks like there's a separate opt-out mechanism for the non-ChatGPT parts of OpenAI (the GPT-3 APIs etc) https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance
How your data is used to improve model performance | OpenAI Help Center

Learn more about OpenAI's data usage policies for our API, ChatGPT and DALL-E

Show threadMarc PabstJan 3, 2023
@dingemansemark Sounds highly illegal under the GDPR (at least in the EU).
Show threadjonny (nonvenomous)Jan 3, 2023

@dingemansemark #AltText4You - text in above image:

  • Who can view my conversations?

    • a. As part of our commitment to safe and responsible Al, we review conversations to improve our systems and to ensure the content complies with our policies and safety requirements.

  • Will you use my conversations for training?

    • a. Yes. Your conversations may be reviewed by our Al trainers to improve our systems.

  • Can you delete my data?

    • a. Yes, please follow the data deletion process here

    https://help.openai.com/en/articles/6378407-how-can-i-delete-my-account

  • Can you delete specific prompts?

    • a. No, we are not able to delete specific prompts from your history. Please don't share any sensitive information in your conversations.

  • Can I see my history of threads? How can I save a conversation I've had?

    • a. No, a view of your conversation history is not possible at this time, but this is a feature we are looking into.

    How can I delete my account? | OpenAI Help Center

    I'd like to delete my account, but I can't find any way to do this.

    Show threadAn AgoraJan 3, 2023
    @jonny
    https://anagora.org/AltText4You
    Node [[alttext4you]] in anagora.org

    The Agora is a crowdsourced distributed knowledge graph: anagora.org.

    Show threadJelle ZuidemaJan 2, 2023
    @evanmiltenburg "crowdsource appropriate responses". I think you're in fact exactly describing the crucial innovation (#RLHF) that has made #chatgpt catch the global imagination so much more forcefully than #gpt3.
    Show threadEmiel van MiltenburgJan 2, 2023

    @wzuidema I guess, but it's a double edged sword if we don't know what the training data looks like. Evaluation is difficult enough as it is.

    (But I guess people are creative enough to always find new ways to show how models fail.)