Ask HN: What are your thoughts on LLMs and user privacy?

5 points eniz | 10 comments | 15 May 25 11:08 UTC | HN request time: 1.576s | source | bottom

I'm increasingly curious about user privacy when it comes to hosted LLM usage. Many popular LLM apps and APIs require sending your prompts, messages, and potentially sensitive information to remote servers for processing—sometimes routed through multiple third-party providers.

- How much do you trust major LLM providers (OpenAI, Anthropic, Google, etc.) with your data?

- For those working on or deploying LLM applications, what approaches do you take to maximize user privacy?

- Do you think end users are generally aware of where their data is going and how it's being used, or is this still an overlooked issue?

I'd love to hear your perspectives, experiences, or any best practices you recommend on privacy when deploying LLM-powered use cases.

1. JohnFen ◴[15 May 25 14:02 UTC] No.43995206[source]▶
> How much do you trust major LLM providers (OpenAI, Anthropic, Google, etc.) with your data?

I have zero trust in these companies on this count, and that's the main reason why I avoid using products that incorporate "AI".

replies(1): >>44001323 #
2. Bender ◴[15 May 25 17:39 UTC] No.43997391[source]▶
I think combining LLM's with big-data was the most genius way to extract highly secret and sensitive information from individuals, intellectual property from companies, secrets from governments and military operations, pillow talk from executives and politicians. I could not have come up with a better way to burn it all down given all of these secrets will "leak" at some point. I fully expect it to become a master of manipulation especially for those that think their LLM is their significant other. Social media manipulation algorithms will be supplanted by AI.
3. scarface_74 ◴[16 May 25 02:31 UTC] No.44001323[source]▶
But you trust Google, Gmail and a host of other online service providers? How do you know that Google isn’t using your email to train? You either have to trust your service providers or don’t use them.

This is like the early days when people didn’t trust buying things over the internet

replies(1): >>44004079 #
4. fosco ◴[16 May 25 04:12 UTC] No.44001779[source]▶
I think it’s terrible, I asked one to find a picture I thought was copied to me from somewhere was a quote not knowing it was made by that person.

The LLM uploaded to internet and “found” the exact picture stating it didn’t exist anywhere else, that source is still viewable on the internet despite me immediately submitting a request to remove. It’s not private but I felt like I’d been had.

5. throwaway519 ◴[16 May 25 11:30 UTC] No.44004079{3}[source]▶
> This is like the early days when people didn’t trust buying things over the internet

If you like bad analogies, why not do car analogies? For example, at least this one's accurate:

I wouldn't trust Sam Altman any more than a used car salesman. The only difference is, Sam Altman's persuaded me to pay him to sell me as a the product.

replies(1): >>44004289 #
6. scarface_74 ◴[16 May 25 11:54 UTC] No.44004289{4}[source]▶
But you’re already trusting Google I assume with your data or Apple who is training AI models. There is no diffeevcd
replies(1): >>44006629 #
7. throwaway519 ◴[16 May 25 15:26 UTC] No.44006629{5}[source]▶
Victim to Google indeed if I email someone at GMail.

I do not use Apple devices, and I do not use Google services. While certainly not the majority, I don't this is that an unusual position on HN.

replies(1): >>44007293 #
8. scarface_74 ◴[16 May 25 16:26 UTC] No.44007293{6}[source]▶
You think the majority of HN is that much different than the rest of the population that uses either iPhones or Android devices running stock Android?
9. greyjoyduck ◴[18 May 25 09:33 UTC] No.44020112[source]▶
It's kind of a nightmare as of right now considering they are basically processing any data they can get their hands on