hhmx.de

· Föderation EN Sa 02.11.2024 01:56:31

@lrhodes @nazgul The less iffy option sounds more likely to me - he does specifically mention the caption being the bottleneck in terms of time, and preemptively uploading photos seems like a massive waste of resources for Instagram unless they can reasonably anticipate an intent to post them - but wow is that still not great.

Föderation EN Sa 02.11.2024 02:04:38

@DamonWakes @nazgul I don't know that I'd put it past them. If nothing else, unposted photos might still make useful training data, especially if they can be linked by metadata to photos that have been uploaded and captioned. But large transfers of that sort would be easier to spot if you're watching for infosec leaks, and I would expect most phones to meter access better than that. But maybe not.

Föderation EN Sa 02.11.2024 02:38:49

@lrhodes @nazgul The video mentions "choosing 512 by 512" which sounds like way too small an image size to be at all recent, which suggests that gathering training data wouldn't have been a concern when that decision was made. I wouldn't absolutely rule out a less charitable interpretation either, though - especially nowadays, since they've likely become more grabby for user data rather than less.

Föderation EN Sa 02.11.2024 04:57:40

@lrhodes @DamonWakes @nazgul THIS. Why *wouldn't* they send your photos into their AI maw if they had them? Speaking of class action... IANAL but I would argue that taking it before you finalize is like assuming agreement on a contract not yet signed.

Föderation EN Sa 02.11.2024 11:14:27

@vansice @lrhodes @nazgul At the point this particular decision was made, I don't think the AI maw existed: mastodon.sdf.org/@DamonWakes/1

Föderation EN Sa 02.11.2024 07:06:11

@DamonWakes @lrhodes @nazgul Taking your photos with or without your permission is a massive money maker for these guys. Your photos contain meta data that is more than images. You bet they’re doing it. They aren’t making billions without selling your data.

Föderation EN Sa 02.11.2024 07:18:14

@Catawu @DamonWakes @lrhodes @nazgul Chances are good given Meta's history of privacy violations that they are not only harvesting meta data but imagery to populate their own generative AI.

The more image data it has from the real world oriented by the meta data, the more realistic the output of its AI, close to 3D 360 degrees.

Meta could replicate users' lives with this given enough data from around each user.

Föderation EN Sa 02.11.2024 10:07:20

@DamonWakes @lrhodes @nazgul you no read term and conditions?

Föderation EN Sa 02.11.2024 10:38:36

@DamonWakes
That's how I understand it as well: by the time you're writing the caption, you're already using instagram. So I don't think there's a risk of uploading the kid's picture or a photo of a contract.

Still iffy though, esp. if the button is called 'upload'. Downright misleading.

It's like when Facebook was recording the text of the post while you were writing it, which was ending up on their server even if you decided not to send it eventually.

@lrhodes @nazgul