Do you trust OpenAI, to not train on your data?
If you either just use their API or playground, disable the History in the ChatGPT UI or make a privacy request on https://privacy.openai.com then OpenAI states, they won't train on your data. Can you trust a company like OpenAI to actually not train on the data?
Yes, there is no upside for them to explicitly say you can opt out of training using these methods if it's not true. They are already facing copyright infringement lawsuits, and clearly more will come, so if they were lying about this, it would come out in discovery, which would be devastating.
So, basically I'd say you can trust them about this because it would be very much against their interest to lie.
I wonder this myself, also with document attachments and multi modal input (e.g. voice). I also wonder how much they really train on the chat history. I have it disabled but maybe I am too cautious and I am missing out on convenience.