asnelt

> It introduces statistical regularities, but all RNGs introduce statistical regularities.

Agreed.

> So the question is on average are these statistical regularities better or worse than those introduced by the alternative,

Indeed. This is an empirical question. That is what the article is about, for a specific setting.

> ... and the answer is no, if implemented properly.

I don't agree with that. This is not about implementation. It's about how the statistical regularities that are imposed on the full output distribution affect that distribution. There is a change - by construction. That change can be good in some situations and bad in others. The authors claim that it is mostly bad in the setting that they investigated. This looks like a fair statement.

The conditional distribution is the same. The joint distribution of the full output does change with watermarking. Just look at the diagram in the article: there is a loop back to the seed once a token has been generated.

Changing the distribution is the whole point: it introduces statistical regularities that can be detected.

Me too. It seems to be the only one that does not litter the sdcard directory.

I hear you. This is exactly the reason why I have a digital signage display rather than a Smart TV.

> Can we take this as an official statement that Nvidia supports local models?

The first reply on the X thread is by Gerardo Delgado, who is "Sr. Director of Product - Local AI, Creators, Developers @NVIDIA" according to his profile. In his post, he says "My team's goal at NVIDIA is to grow Local AI."

I couldn't immediately find whether GrapheneOS grants READ_PRIVILEGED_PHONE_STATE to Google Play. It might very well be that the GrapheneOS sandbox spoofs a fake IMEI, and I do hope so.

In any case, my parent comment was meant for stock Pixels, as mentioned by armadyl further up in this thread.

Even with a burner sim, there is the International Mobile Equipment Identity (IMEI) number, which is tied to the phone, and is known to all apps with the android.permission.READ_PRIVILEGED_PHONE_STATE permission.

The outcome of this vote is not set in stone. If the situation changes, there can be another general-resolution vote to adapt to the new situation. I think the result of this vote is appropriate for the current degree of automation when used responsibly.

Exactly. Just tweaking the order of suggested models for a search query could already have a huge impact on model popularity.

> AOSP has a massive amount of non-Google companies and individuals working on operating systems based on it.

based on is the important part here. Yes, AOSP would survive without Google. But as long as Google is in control of AOSP, it controls the direction in which AOSP and all downstream projects are headed. Because of breaking changes in AOSP, there is a cost in maintaining downstream changes. So downstream projects cannot veer off too far, short of a hard fork.

> You're misinformed about security preview patches. It has nothing to do with Google being hostile.

Does it not? From the 2025092500 GrapheneOS release notes: "We strongly disagree with broadly distributing patches to OEMs 3-4 months before the official publication date. It further delays getting patches to users and sophisticated attackers will have no issue getting the patches from one of many people at Android OEMs with early access."

Sounds pretty hostile to me.

That's an interesting video. If the skyscrapers were flickering just a little faster, the sky would be the color of a television tuned to a dead channel.

> If it bears the mark, it is likely generated (or someone deliberately made it look generated).

One could even say, the mark is load-bearing.