Some are even multimodal LLM. You can transcribe audio using Gemma 4 for example. While it is not the primary purpose of an LLM it has the advantage that it can do more than speech, like describing other sounds and that context a LLM knows that a simpler TTS engine does not know can prevent transcript errors. Every silly error you see in automatic captions that is obvious nonsense can be caught by an LLM. The subtle ones stay subtle, of course.

Or just a ban for VPNs that don't store customer data? For age verification only, of course ...

Probably everything above a certain rate of floating point operations. Which means your future 6090 card would probably need to have such a chip.

People understanding that snaps sucks and people using Ubuntu have probably very little overlap ;)

Do you know how many people never learned to touch type? I prefer a keyboard very much and are way faster with it. But most people type veeeery slow.

unattended-updates should only install security update (except Ubuntu did have some special ideas again ...)
The problem is snap automatically updating if you want it or not. I think without low level tampering (disabling snap basically) you can only pause updates for a limited time, but not disable them. That feels like iOS/Android (yeah, also Android. Try to disable Google play service updates ...) bullshit.

All current (and all actually working) speech recognition is neural networks and by this clearly in the category "full AI". In principle the old pattern matching approaches are also AI (which is a way broader category than most people think).

Why stop at the facial recognition? You can't control that as passenger anyway. Do they have it, do they not? Proof it!
Let's make them remove the cameras. Surveillance does not start with the processing, but with the recording.

You are allowed to copy these for your friends (but not for selling or public sharing). You are not allowed to copy video games.

People may say "63% fail rate is bad!" but new AI benchmarks are designed so current models fail, because when all models are in the range 85-90% of old benchmarks, the signal/noise ratio of the benchmark is low. So when this happens you introduce a benchmark that challenges current models to observe future models becoming better.

but perpetual licenses together with a medium that worked without internet access. If they would have sold you the game, you would have been allowed to copy it.

I don't really oppose the idea per se, but I don't think it is realistic. Neither the idea itself, nor realizing it with the current US government.

People warned you about Adobe since about 10-15 years. Don't act surprised, you accepted each time their updated ToS instead of migrating to alternatives. It's not "How can I migrate urgently right now!" but "Why did you not migrate ten years ago?"

When should we collect the tax? The next startup will say "We're worth $1000, collect your $500 one time tax right now" to avoid paying the billions later on.

Indeed it is. Don't confuse AI with NN.

https://www.researchgate.net/f...

Most spam filters are even ML. Even the bayesian filters 20 years ago had a learn spam / learn ham function. And yes, bayesian filters are ML.