Hacker News Reader

Comment by orochimaaru | Hacker News Reader

Why is this surprising? They’ve always done this. In fact I’d be surprised if they didn’t do this. Fwiw - llama is free to use. So I guess it’s a good enough return.

I don’t use Facebook. I’m not sure if they can peek into WhatsApp messages.

View on HN

> Why is this surprising? They’ve always done this. In fact I’d be surprised if they didn’t do this.

This is such an unconstructive attitude. This is the first time they have publicly admitted it.

It's the first time they've publicly acknowledged what specifically was used in the training set for their LLMs. It's NOT the first time they've said that that data could be used for research. The presumption was that they had used the data for their LLMs.

Maybe it is surprising to some that they this particular research used that data, but it really shouldn't be to anyone aware of how LLM development is done.

Yeah, to me it's part of what press critic Jay Rosen calls The Church of the Savvy. It's kind of a performative cynicism where one tries to gain status by appearing so smart that you're above it all. One can do it with pretty much anything, which ironically means it demonstrates very little actual smarts.

To me it's related to the sort of person you'll see who on every startup failure says how they knew it wasn't going to work. Which again doesn't require particular smarts; most startups fail, so predicting failure doesn't take a genius. What I think is much more interesting is spotting a problem before it's obvious and naming it in advance. Or better, fixing it early on. That doesn't get you many internet points, though.

was it supposed to be surprising?

it's still good to have confirmation of these things that we all assume to be true.

> Why is this surprising? They’ve always done this.

Ah, well, that makes it alright then. Move along, people, there’s nothing to see here. What’s that? What do you mean you didn’t know about the company slurping your data for their own personal gain, or knowingly poisoning you¹ and selling you defective deadly products²? Didn’t you read that one forum comment by a random person that one time? You shouldn’t be surprised. They’ve always been evil, so what can we do other than cross our arms? All hail our corporate overlords!

> I’m not sure if they can peek into WhatsApp messages.

But someone knows. And if you use WhatsApp and get screwed later, wouldn’t you rather not having your concerns dismissed?

I don’t use WhatsApp, but do have an anecdote. I have a friend who just started using Instagram and purposefully follows no one. She doesn’t have Facebook. Recently a friend recommended to her a specific product on WhatsApp and then she started getting ads for it on Instagram.

¹ https://www.sydney.edu.au/news-opinion/news/2024/05/02/how-c...

² https://www.decof.com/documents/dangerous-products.pdf

They did get sued successfully recently by an AG for the autotag faces in photo feature, so maybe another lawsuit is in the cards.

Need to test how good facebooks sanitization was, maybe you could find some PII in llama responses with the right prompt.

I'd assume that was one thing they were extremely diligent about ironing out, doing this at scale.

What is that assumption based on?

There have been numerous scandals over the years regarding their lack of care when handling personally identifiable data.

Meta is not who I look to for being extremely diligent with data.

Because open source, publically-released LLMs regurgitating specific user data is enough of a strategic risk to the entire effort that it would have been #1 on the test cases.

LLM training is also apple and oranges with regards to how they handle data in the course of normal processing.

yeah I'm surprised that anyone didn't see this coming. The amount of times i've heard, "this is all our customer data, perform <xyz transformation> on it," I would have about $5.

Reply on news.ycombinator.com