Hacker News Reader

Comment by ilamont | Hacker News Reader

It's not a change confined to Twitter; it's a glimpse into what could be the new normal for the internet.

You can bet Google/Gmail/YouTube, Amazon, Microsoft, TikTok, and every other Internet platform that works with user-generated content will soon do the same ... if they haven't done so already.

View on HN

GMail's already been using your data for classification, data mining, etc. other AI purposes. I would be legitimately surprised if they're NOT using processed GMail data (potentially removing sensitive data, etc.) in training their LLMs or other AI projects.

Yup. To be clear, that's with the free version though.

If you pay, corporate versions of Google Workspace won't train on your data. That's very much by design, since companies don't want anything internal ever being exposed.

But with the free version, that's part of what you're "paying" for it to remain free.

Could you share a source? I'll immediately switch off my gmail onto one of my google workspace domains, if so.

https://workspace.google.com/blog/identity-and-security/prot...

https://9to5google.com/2023/07/03/google-privacy-policy-ai-t...

Although judging from that, it sounds like Google may have changed policy so that even private data in the free versions is no longer used for AI training.

They stopped targeting ads in free Gmail based on your e-mail contents years ago because of the bad press. So maybe they've stopped training AI on free Gmail/Docs data out of similar precaution, now that LLM's are everywhere in the news.

They'd certainly use it to train their spam classifiers, but using that data in a generative model shared with other users would risk information leaks, so they wouldn't do that.

They've been doing it for years. Everything you put on their servers is a source for them to train on. At FB Messenger in the mid to late teens, the suggestion and auto-reply models were trained on the entire corpus of unencrypted messages sent between users.

One of the big reasons why I went with Proton Mail for whitelabeled email service when I stopped hosting my own. Nothing to scrape on the server side.

The absolute irony of Google doing this.

Reply on news.ycombinator.com