I.e., to what extent does search activity reveal confidential information to an external entity (Google) that has no right to that information, nor any restriction on further sharing it?
E.g., could Google legally sell EPA search queries to companies bidding on competitive EPA contracts?
Either way, is there a clear line between using Google and using ChatGPT w.r.t. information leak?
I think Google and OpenAI are equally likely to collect/use/sell any information you give them, so in that sense the risk is the same, but you are incentivized to give way more information to OpenAI, at least if we're comparing search and ChatGPT. Nobody is copy/pasting an entire document into a Google search (it's useless and also it doesn't let you), but you are encouraged to do that in ChatGPT.
Obviously you could also use Google for email and document storage, potentially giving them all that data, but I'm sure the EPA has separately assessed the risks of using cloud providers for those things.
This. I think the crux of the issue is that people have never found that copy-pasting blobs of confidential data was rewarding because, comparatively to GPT, pasting in google is, how can I say, a bit dumb.
Now we're seeing improvements to search where pasting large bodies of text provides actual value and it just needs to be reiterated that copy-pasting data into a website gives said website access to the data, so don't do that. Although the framing of this discussion has been around 'pasting into AI chats is bad' it really should be "don't leak confidential information to websites, chat GPT included".
I.e., to what extent does search activity reveal confidential information to an external entity (Google) that has no right to that information, nor any restriction on further sharing it?
E.g., could Google legally sell EPA search queries to companies bidding on competitive EPA contracts?
Either way, is there a clear line between using Google and using ChatGPT w.r.t. information leak?