Fine-tune ChatGPT with & data security

Risk and safety
1

What steps can organizations take to ensure data security and privacy when fine-tuning ChatGPT to improve its natural language generation abilities, and how can they ensure ownership and control over the resulting model? Additionally, are there any limitations or considerations organizations should be aware of in terms of OpenAI’s rights to the fine-tuned model?

  1. Was the training data saved by OpenAI during the fine-tuing process of the ChatGPT model?
  2. Does OpenAI retain exclusive ownership of the fine-tuned model, or are there other parties who may have ownership rights?
  3. Does OpenAI have the right to use the fine-tuned model to improve future versions of ChatGPT, or are there limitations on its use?
2

Hi @bengadisoufiane

Welcome to the community.

ChatGPT aka gpt-3.5-turbo does not support fine-tuning yet. Only base models ada, babbage, curie and davinci support fine-tuning

1 Like
3

Sorry, but you didn’t answer my questions : my questions are about data security

4

Yes, but you said @bengadisoufiane, precisely:

and what @sps has correctly pointed out to you is that you cannot fine-tune any “ChatGPT” model.

So, since you cannot fine-tune a ChatGPT model, there cannot be any “data security” concerns about a fine-tuned ChatGPT model.

HTH

:slight_smile:

1 Like
5

But what about data securtiy and ada , babbage , curie and davinci models ?

  1. Was the training data saved by OpenAI during the fine-tuing process of the ada model?
  2. Does OpenAI retain exclusive ownership of the fine-tuned model, or are there other parties who may have ownership rights?
  3. Does OpenAI have the right to use the fine-tuned model to improve future versions of the ada model, or are there limitations on its use?
6

We assume so.

This is question for OpenAI, not for a community of software developers. Try:

support@openai.com

Yes, if you read the terms of service on the OpenAI web site, OpenAI and their contractors, etc, have full rights; and you, the user, have no warranty, implied or otherwise, your data will be private.

You would get better answers, I think, reading the OpenAI Privacy Policy:

HTH

:slight_smile: