If hypothetically targeted ads would be banned, there would be the same advertising budget as before, the problem is everyone would have to stop at the same time. It's a prisoner's dilemma.

There are also quite a few articles that targeting doesn't even work that good. But put ads without personalization on your page and you earn less. Offer ads with less personalization and people choose the companies that offer better targeting. Everyone needs to do it, because everyone else does it. No matter if it is effective or not.

I mean music is not far from language, but LLM can still not generate music. That are completely different architectures.

They say
> [The] consumption of these tracks is still very low, "between 1-3% of the total streams

and later say
> Songs tagged as AI-generated on Deezer are automatically removed from algorithmic recommendations and not included in editorial playlists.

I wonder if there may be a causal connection.

The only way to stop tracking and datamining is to kill the monetization. They will collect what they can find, but if nobody buys the data, they will eventually stop. So next ban personalized ads.

Hexchat did this for ages. Free on most OS, commercial on Windows because building for Windows is more annoying.

Why not, if the app is open source.

Why should they fight in court for a case they would clearly lose? They did it because they think they won't be found or can't be persecuted because of the legislation they are in. Why should they defend themselves in an US court for something that is illegal in the US?

Sure, AI is to blame. Totally believable.

By reversing the timeline, people cannot start scrolling where they left off the last time. This means instead of "I continue reading when I have time" they feel the urge to keep up with their feed, because they otherwise will miss out on the things that happened between now and the hard to find point where they stopped reading. Add "algorithmic" sorting to the mix and it becomes even harder to keep reading at your own pace without missing out on something.

I thought the last real bird died in 2001?

The question is, if the asshole moves were just arbitrary because of a lacking objective, or actual minima. If they were arbitrary you add the problem the the loss function for the next model and have a good chance for a better solution. If the AI only created a plan, the even simpler alternative would be to generate 100 plans and select the non-asshole ones. There is no need to automate A to Z just because AI is part of the pipeline. Add as much human influence as you need for a good result (including worker health).

There would be no sense in quitting for that. Their point is: We want to quit X. Now that we don't need it anymore, we finally can.

Predicting a bubble in 1-2 years ... that's not how bubbles work. If it is a bubble, nobody knows what's in 2 years. Bubbles are a short-term effect.

Opera did it like 20 years ago without extension.

put your own signing key in the UEFI. You only need Microsoft if you want to be able to verify unknown software. If your Ubuntu is signed, you do not know if by Canonical or not, if you don't have a trust anchor. If you sign your own kernel, you know what key you put in the UEFI and everything is fine.