Live data from Hacker News

Google denies 'misleading' reports of Gmail using your emails to train AI

theverge.com

51–58 of 58 posts

Re: Google denies 'misleading' reports of Gmail using your emails to train AI

#51
post #4

Poor AI, reading all those ad emails...

I am wondering can we use LLMs to semantically encrypt our emails so that if I am talking about my startup strategy, to the person snooping or NSA it will appear as if we are talking about recipes.

We're proposing semantic steganography using LLMs as encoder/decoder pairs where startup strategy discussions appear as recipe exchanges. Unlike traditional crypto, security emerges from semantic complexity rather than mathematical hardness - the LLM maps between concept spaces (e.g., "fermentation time" ↔ "development cycles") using its world model. Both parties share a seed phrase that deterministically generates the same bidirectional mapping, eliminating key exchange over insecure channels. The core insight: natural language is already an encoder (concepts → symbols), so we're just adding a second semantic layer that looks like normal Layer-1 communication to observers. Main challenges are LLM non-determinism requiring error correction and the tradeoff between information density and plausibility. The approach essentially weaponizes the LLM's semantic understanding to create a regenerable codebook rather than storing/transmitting it.

Re: Google denies 'misleading' reports of Gmail using your emails to train AI

#52

When you disable all of these features, eventually it turns off email categorisation. At first this was annoying to me because it’s obviously a very good feature. But the last few weeks have been quite revealing: I’ve been receiving and unsubscribing from tons of emails I had no idea I even received regularly, because categories buried them away. I wonder how much of newsletter marketing (and paid email marketing) is…

Exactly the same for me. I unsubscribed from more newsletters in the past few days then in the last couple years.

Re: Google denies 'misleading' reports of Gmail using your emails to train AI

#53
post #6
post #2

> Enabling the feature in Workspace says that “you agree to let Google Workspace use your Workspace content and activity to personalize your experience across Workspace,” according to the settings page, but according to Google, that does not mean handing over the content of your emails to use for AI training. Google be like: "trust me bro"

User-facing software is full of language like that these days and I find it really frustrating, because it never helps answer the questions attentive people actually have, like will that mean my emails get dumped into the next Gemini training run?

Maybe my brain has rotted from paying a bit more attention to privacy language than the average person, but IMO in this case it's fairly clear to me that using your activity “to personalize your experience across Workspace” does not mean using it for training Gemini. The “personalize” means it's for training the recommendation/categorization systems for you, like which emails get marked as "important" or not (the settings at https://mail.google.com/mail/u/0/#settings/inbox).

(In some very broad sense I guess critics could call this “training AI” as there's an ML system somewhere whose parameters associated with your account get updated, but I think we can all agree this is not what we think of as “training AI”, i.e. going into a cross-user dataset for training Gemini or whatever.)

(I guess what Google should do, and should have done years/decades ago, is create a fixed set of categories of how your data can be used (aggregate statistics, training Gemini, personalization…) and use the same language across products, legal, everything.)

Re: Google denies 'misleading' reports of Gmail using your emails to train AI

#54

When you disable all of these features, eventually it turns off email categorisation. At first this was annoying to me because it’s obviously a very good feature. But the last few weeks have been quite revealing: I’ve been receiving and unsubscribing from tons of emails I had no idea I even received regularly, because categories buried them away. I wonder how much of newsletter marketing (and paid email marketing) is…

What’s annoying is some newsletters don’t respect the unsubscribe action and continue pumping the spam.

Re: Google denies 'misleading' reports of Gmail using your emails to train AI

#55
post #53
post #6

Earlier quoted context omitted.

User-facing software is full of language like that these days and I find it really frustrating, because it never helps answer the questions attentive people actually have, like will that mean my emails get dumped into the next Gemini training run?

Maybe my brain has rotted from paying a bit more attention to privacy language than the average person, but IMO in this case it's fairly clear to me that using your activity “to personalize your experience across Workspace” does not mean using it for training Gemini. The “personalize” means it's for training the recommendation/categorization systems for you , like which emails get marked as "important" or not (the se…

I interpret it that way too, but my bigger problem is in explaining it to other people. I can't credibly point nervous users at that language and say "what this actually means is X", because there's too much vagueness and wiggle room built into the language that the companies publish.

Re: Google denies 'misleading' reports of Gmail using your emails to train AI

#56
post #2

> Enabling the feature in Workspace says that “you agree to let Google Workspace use your Workspace content and activity to personalize your experience across Workspace,” according to the settings page, but according to Google, that does not mean handing over the content of your emails to use for AI training. Google be like: "trust me bro"

It sounds like you have some serious knowledge gaps about AI. It's perfectly normal to use AI on a dataset without incorporating that dataset into training a new AI model. If you download a free model and run it offline on your data, your data doesn't get magically incorporated into the model on the site you originally downloaded it from.

This is the same thing, and it's backed by a contract and the threat of lawsuits from the many businesses using Google Workspace.

Re: Google denies 'misleading' reports of Gmail using your emails to train AI

#57
The fact that Gmail doesn't allow you to customize or expand your categories beyond the 4 it gives you frustrate me beyond belief. That, and the fact that there are countless examples of email sources that no matter how many times you move it to the category you want and tell it to "send all emails from x to category from now on," it continues to fail to filter them unless you make your own filter. And then managing the filters is a pain in the ass with a UX that hasn't been updated for 15 years, and also their labeling system is stupid because selecting a parent label only shows you emails with that EXPLICIT label instead of including all children labels.

Re: Google denies 'misleading' reports of Gmail using your emails to train AI

#58
post #5

The only reason I am still using gmail is due to choice paralysis. I do not know which email service to choose and pay for. I do not like Proton. Is Fastmail the way to go? There is also the German one posteo. Should I just use Apple's mail? I'm taking suggestions if you have anything to share.

I've been on Fastmail with my own domains for 4 years now, having been fully Gmail since 2004. The transition was seemless, the apps are solid and I can't imagine ever going back. The only option I'd consider would be full self-hosting, but I really can't justify the effort to maintain the one service where I need very high uptime.
Post reply on HN