Remix.run Logo
▲ nostromo an hour ago

Quick public service announcement: both Anthropic and OpenAI allow you to turn off allowing them to train using your data.

▲simonw an hour ago | parent | next [-]

Doesn't help if someone you sent information to pastes it into the chatbot under their not-opted-out account.

▲TeMPOraL an hour ago | parent [-]

Well if you have some specific confidentiality requirements for some information, don't send it to people who you aren't 100% sure will respect those requirements.

And if you're publishing something in general, someone feeding it to a chat bot is really low on the scale of things to be concerned about.

▲minraws 35 minutes ago | parent [-]

What if someone has mail connected and it reads fhe emails this nonsensical you cant share even bits of i formation now huh?

Same issue with phones they transcibe and allow summarizing calls who knows which models they use.

▲TeMPOraL 28 minutes ago | parent [-]

That's peak of AI psychosis, you know.

And if anyone actually sold such phones, I'd be first to buy it. LLMs are godsend for dealing with daily communications, but this process is still mostly manual on end-user devices.

▲zulln 16 minutes ago | parent | next [-]

I mean, even iPhone has built-in transcription [1]? Pocket [2] offers it for any phone through a contact microphone (so records the vibrations of your phone).

No idea what is their policy on training on calls, but as a concept is sure exists already, while you seemed to indicate otherwise.

[1] https://support.apple.com/en-gb/guide/iphone/iph57c6590e9/io... [2] https://heypocket.com/

▲tavavex 14 minutes ago | parent | prev [-]

Let's be clear here. The 2026 redefinition of "AI psychosis" = "being concerned about the data you intended to be private being seen by enormous megacorporations"?

There's no barriers whatsoever to getting a phone that will spy on you by default. Lots of Google apps will happily do it. The spying doesn't obligate them to provide useful features in return, you're just feeding your data into their datasets.

▲contubernio an hour ago | parent | prev | next [-]

Professional mathematicians just don't believe this.

▲BigTTYGothGF an hour ago | parent | prev | next [-]

Why would anybody possibly believe this?

▲dw_arthur 31 minutes ago | parent [-]

Anthropic and OpenAI stole all the information from the Internet yesterday, but this is today.

▲mirekrusin 40 minutes ago | parent | prev | next [-]

And what's your interpretation? Ie. if they convert/summarize your data and train on that – is it "your" data or "theirs" now? What is their interpretation?

▲coldtea 26 minutes ago | parent [-]

Does it matter? Only matters if a court will ever fine them enough to care, which they wont.

▲irthomasthomas 28 minutes ago | parent | prev | next [-]

Except Fable

▲prometheus1992 an hour ago | parent | prev | next [-]

can someone flag this bs please.

▲jotux an hour ago | parent | prev | next [-]

    [ ] Check this box and we'll *totally* not use your data for training
They might as well have this box as well

    [ ] Check this box if you'd like to preemptively be paid $25 as the result of a future class-action lawsuit because we just used the data anyway
▲helterskelter2 43 minutes ago | parent | next [-]

$25 would be a record for me. $5, tops.

▲jotux 32 minutes ago | parent [-]

$25 worth of fable tokens!

▲xp84 an hour ago | parent | prev [-]

Even though I'm not as cynical as you on this, this is S-tier humor. All privacy choices everywhere should be framed this way. I wonder if I framed all of the notices of data breaches and slap-on-the-wrist settlement privacy violation notices I've ever gotten, if I could cover the walls of a whole room with the frames at this point.

▲coldtea 25 minutes ago | parent [-]

Cynical or realistic? Major tech companies have been caught doing that hundreds of times.

▲trinari an hour ago | parent | prev | next [-]

So you trust the biggest information hoovers of our time to always respect such settings?

▲WarmWash an hour ago | parent | next [-]

Yes, because they need enterprise money more than anything.

▲afavour an hour ago | parent | next [-]

Enterprise accounts are a whole different ballgame though. Not at all implausible they’d do one and not the other.

▲xp84 an hour ago | parent | next [-]

It still would be toxic to corporate decision makers if they were caught dishonoring this control on one type of an account. If they're willing to lie and cheat on one tier of account, what guarantees they won't lie and cheat on the other tier next quarter. Especially with Anthropic being a public company now, that pressure would certainly be there if their corporate culture were comfortable with this type of fraud.

▲afavour an hour ago | parent | next [-]

But didn’t they already train their AI systems by violating copyright, vacuuming up any and all material they could find? Hasn’t that bridge already been crossed?

▲xp84 36 minutes ago | parent | next [-]

There's a difference between "These guys probably downloaded a few hundred thousand books on BitTorrent in order to get this smart" and "These guys will train on my company's internal communications and thus our knowledge and wisdom and trade secrets will be made available to others."

Tangent: I personally literally don't care that all these LLMs trained on a ton of copyrighted material. It is in my humble opinion a long-awaited dose of karma for the copyright lobby, whose own abuses over the decades are numerous. It has basically restored balance to a playing field where the public has had no rights and (mostly) publishing conglomerates have held all the rights and privileges.

▲TeMPOraL 43 minutes ago | parent | prev [-]

No, except in the minds of people who suddenly became copyright maximalist out of envy, and adopted a "dog in the manger" attitude.

▲_aavaa_ an hour ago | parent | prev [-]

Such privacy checkboxes always seem to suffer from a string of bugs that causes them to reset to the "suck all my data" option. It's really quite strange.

▲WarmWash 10 minutes ago | parent [-]

API usage has no checkbox to accidentally click.

▲rowanG077 13 minutes ago | parent | prev [-]

They'd lose all credibility if they would not respect the setting of non-enterprise users. Like people are just papering over "LOL, do you actually think they respect that?" The consequences if they get caught are staggering, so yes I actually believe Anthropic is not that stupid to destroy their own business.

▲marricks an hour ago | parent | prev [-]

I think they need to prove relevancy and potential for future disruption more than anything to keep IPOs and investments big.

That means stealing leading researchers info is probably great for them, whether it's intentional or not.

▲iterateoften an hour ago | parent | prev | next [-]

Also they can adhere to the settings of no training and still use your data otherwise.

They have all the best ideas flowing through there servers in real time as they are hatched around the world. It would be tempting to have top scientists accounts posted to internal slack channels to monitor new discoveries.

▲ebilgenius an hour ago | parent | prev | next [-]

yes

▲TeMPOraL an hour ago | parent | prev [-]

Yes, also why is that even a topic here?

Don't post confidential information where there's a risk of confidentiality being broken. That includes posting to people whom you aren't 100% sure of know and uphold your confidentiality requirements. Adults usually handle this by contracts with penalties that include financial damages, loss of employment, or prison time.

▲coldtea 29 minutes ago | parent | prev | next [-]

Quick public service announcement: both have shown that they could not care less what we turn off, or even for the legal slaps on the wrist they'll get for saying "oops, turns out we did read them".

▲usef- 8 minutes ago | parent [-]

Where has the that happened with user data? Or do you mean the scraping of public data / books etc?

▲flir 27 minutes ago | parent | prev | next [-]

Encrypting your email might make intelligence agencies look at you that much harder.

▲iterateoften an hour ago | parent | prev | next [-]

OK, then they just have a smaller LLM summarize everything you send them and train off the derivative.

see? easy peasy for them to adhere to "no training".

I wouldn't doubt OpenAI and Anthropic, etc all have conversation or message categorization like they do for flagging bio, weapons, content but use it internally to flag ground breaking maths in real time and kick the convo up the chain of command for further review.

▲xp84 an hour ago | parent [-]

Call me crazy, but I kinda don't really think they need it.

People talk as though all "data" is valuable to AI companies, but ingesting arbitrary bytes into LLM training today, en masse seems like a recipe for degrading its quality.

Even in the domain of mathematics papers. When the frontier models are already really great at this maths stuff, feeding in every blob of text someone grammar-checks or whatever doesn't improve it. You don't know up front if the text is from some crackpot, from a naïve student with minimal understanding, or from the greatest mind of the generation. Though the chances are good that it's not that last one.

It seems like you'd be interested in ingesting published papers that are either from highly respected authors, or from newer authors which the eminent minds have shown respect and praise for. But I just don't buy that if someone pasted in some text, or if Claude Desktop read a draft of this paper in someone's email or whatever, that we'd find that a model had been trained on it.

▲pegasus 12 minutes ago | parent [-]

I think the opposite is the case. Training on the half-formed but ultra-fresh thoughtmatter of humanity allows these models to push the frontier of intelligence. Separating the wheat from the chaff can be done in a similar way they most likely do generally: for example, by looking for a wide range of specialized vocabulary, or using a custom trained decision model etc.

▲233mhz an hour ago | parent | prev [-]

I have a bridge to sell you