How to ask Facebook's Meta to not train its AI models on some of your info | Hacker News Reader

How to ask Facebook's Meta to not train its AI models on some of your info

theregister.com

17 pointsLinuxBender12 comments

I'm surprised I haven't seen my aunt post a Facebook status update like "I do not give Facebook or Mark Zuckerberg permission to use my data, images or messages for AI training. Repost so you don't give Facebook your personal information. I heard from a friend this works"

I suspect they've largely flagged those as spam at this point.

I’ve seen my actual aunt post almost exactly that text multiple times. Someone tried to explain it to her, but alas…

If Meta allows us to opt out of having our data harvested, then doesn’t it imply that we should really be offered the opportunity to opt in?

Then, opt out should be the default.

How can the company explain this?

> How can the company explain this?

That the vast majority of users don’t care and neither do lawmakers.

In my experience, most people do care, quite a lot. The trouble is that they don't understand the issue, and when told what these companies are doing, they don't believe it.

>when told what these companies are doing

lol

The models are not doing what advocates of this kind of setting say they are doing, so it will be nice when the concerns fade away.

Using data to train models is what they are doing.

Storing copies or parts of images in the model so exact lookalikes will be regenerated from the model is what mistaken people say they are doing.

I mean sure it’s possible a model can generate your likeness but it would also be able generate it if you had not been in the training set. And it’s possible atoms can spontaneously assemble into a teapot orbiting the sun but that doesn’t make it likely.

“And it’s possible atoms can spontaneously assemble into a teapot orbiting the sun but that doesn’t make it likely.”

Ah thank you for making me think of Douglas Adams and smile.

> Using data to train models is what they are doing.

Yes, among other things.

> Storing copies or parts of images in the model so exact lookalikes will be regenerated from the model is what mistaken people say they are doing.

I'm not sure why you're saying this as a counterargument to me, considering that I've never made that argument in the first place.

Thanks for clarifying.

I chose what seemed the most likely interpretation of an ambiguity.

If it works for you, think of it as a drawing out of the argument, instead of a counterpoint to you. I’m arguing against the others who think there has been some grievous wrong committed by training models on this data.

The data is not being used in the way alarmists describe things, so it’s on the alarmists to explain why they think there is a problem.

Of all their data practices feeding an AI is the least of my worries. In a way it obscures it. Selling it verbatim and the whole dodgy data broker eco system around FBseems far more problematic

Reply on news.ycombinator.com