[superposeur]

Everyone seems to be unsurprised by this move, but I’m genuinely shocked. What a shoot your own foot business decision. Google, evil though it be, doesn’t post the text of your gmails in its search results because who would consider using Gmail after that? This is the llm equivalent. Am I missing something?

[rs186]

Gmail used to serve ads based on your emails for many years until 2017. https://www.npr.org/sections/thetwo-way/2017/06/26/534451513...

[skylurk]

And in 2010 they made https the default. Different times :)

[einpoklum]

Google mines the bejeezus out of your email, and uses it to any number of ends, including manipulating you into buying things, and also passing your correspondence on to the US government. While this is not the same as outright making your emails universally searchable - training Claude on your emails is also not the same as posting their contents.

And - this behavior of Google's has not been penalized, I'm afraid.

[aleph_minus_one]

> Am I missing something?

I think you do:

According to the article https://www.perplexity.ai/page/anthropic-reverses-privacy-st...

"Enterprise and educational customers will continue operating under their existing privacy protections, as the policy changes specifically exclude Claude for Work and Claude for Education services. These commercial accounts remain governed by separate contractual agreements that maintain stricter data handling standards.

Organizations using Claude through business partnerships or educational licenses can continue their operations without concern for the new training policies affecting their sensitive communications or proprietary information."

Thus, I think your claim

> What a shoot your own foot business decision.

likely does not hold: the non-commercial accounts likely led to Anthropic loosing money, so they are not liked by Anthropic anyway (but are a an "inconvenient necessity" to get people to notice and try out your product offering). With this new decision, Anthropic makes this "free-riding" less attractive.

I bet that Anthropic will soon release a press statement (that exists in the drawers for quite a long time) "We are listening to your concerns, and will thus extend our 'privacy-conscious offering' to new groups of customers. Only 30 $ per month."

[ceroxylon]

> With this new decision, Anthropic makes this "free-riding" less attractive

Certainly not for any users like you and me, it takes two seconds and three clicks to review the new terms and decline chat training. This is more like Anthropic getting easy training from people who are unaware or don't care.

[827a]

Gmail is free. It would still be incredibly bad for Gmail to start publishing the content of free users' emails to Google search.

But also, Anthropic has said that this new policy also applies to their Pro ($20/mo) and Max ($200/mo) plans. So its not free versus not free.

[layer8]

Well, it means that LLMs used for business use cases will be trained on input from non-business use cases of non-privacy-conscious users.

[KoolKat23]

This data is useful for reinforcement learning. All the others do it.

And most importantly, you can just opt-out.

[behnamoh]

Just because all the others do it doesn’t make it right. Many users chose Anthropic exactly because they were not like the others.

[superposeur]

Ok, to be clear, let’s say I’m dumb and accidentally go with the default (I get the color of the opt out button wrong or something). As if there’s a “publish my private emails to the internet” default-on button in email. Then, I use it to edit a rec letter for student X, with my signature Y. (Yes I know this is dumb and I try changing names when editing but am sure some actual names may slip through.) A few months later the next model is released trained on the data. Student X asks Claude what Y would write in a rec letter about X. Such a button is a “wings stay on / wings fall off” button on a plane.

[turnsout]

You can't opt out of the data retention policy.

[shadowgovt]

The LLM equivalent is what Google does do, which is train its spam filters on the contents of your emails coupled to the signal of what human beings flag as spam.

(It was one of the first significant value-adds of GMail: at its scale, Google could create a global-concept understanding of the content and pattern of spam across hundreds of millions of users. That was the kind of Big Data that made it possible to build filters where one could confidently say "This is tuned on all spam in the wild, because we've seen all spam in the wild").

[podgorniy]

What a framing. Like there is exactly a surprise behing all these reactions.