And being cheaper and better at it on top of that.

Indeed. Underestimating China is a really bad mistake at this time.

No. "Hard stuff" takes a lot more time than they have. And probably more skill and education as well. "Cool, flashy" stuff is all they can do.

Yup, one stated reason for the RAT on documents I have seen is to extend the data collection by the FDRs by maintaining power to the sensors and canbuses.

That's how I read it. It should say it has no thrust.

A typical jet turbofan airframe has two engines that each have a generator shaft taking turbine energy and making electrical current. It then has a whole 'nother turbine engine used on the ground and in some other flight legs called the APU; this exhausts out the tail cone usually, and can start engines or provide extra hydraulic power if needed, but is slow to start just like the main engines.

For power loss emergencies, a small spring-loaded fan pops into action super fast, called a Ram Air Turbine or RAT. It can only make enough electrical power to reboot key systems like engine FADECs or avionics, often only on one electrical channel instead of all channels. It's only a turbine, not a thrust-producing fan. It's a pinwheel toy in comparison to the APU and even the APU cannot produce significant thrust.

Yes, it does.

You seem to be deep in delusion. Just one example:

So, your claim is I have made up the term "model collapse"? Well, here is a reference: https://en.wikipedia.org/wiki/...
You think something I made up "on the fly to sound authoritative" has a Wikipedia page with 25 references going back to 2022?

The rest of your statement is of similar quality, i.e. simply crap.

I dare you to find a single posting by me that says LLMs are useless. Or eat your lie.

Yu know what? It looks less likely every day that this use case will really work out. LLMs continue to be as dumb and disconnected as bread and hallucinate on top of that. Training data gets more and more outdated. Scaling LLMs up or iterate things (falsely called "reasoning") seems to do very little compared to the additional effort needed. And more signs that this AI hype is again 98% hype and 2% substance. Like all other AI hypes before.

My prediction is that in 5 years we will be going into the next AI Winter and the usual morons will claim that there was no way to see that coming. Then in 15 to 20 years, we will get the next deranged AI hype. And so on, because people do not learn form history.

I disagree. First indicator is that all AI bubbles so far have popped and usually that was followed by an "AI Winter" because the promises were so grossly exaggerated. It looks to be even worse this time, if anything. Second indicator is that still nobody is making profits from customers with LLMs. And it really does not look like this is going to change anytime soon. Third indicator is that after the initial releases, not much has been accomplished. In fact there are signs of mounting desperation. At the same time, hallucinations have only be mildly reduced, training data is ageing and the legality of the whole piracy campaign that got them the training data is being examined and it does not look like they will simply get away with it.

There are some more. Your mindless cheerleading is really mindless and ignores reality.

Sure. It takes time. ASML needed quite a bit of time too. But China can for sure do it and they will be faster to get there because they know it can be done and what direction to go in.

Some would say that there is not much need for a long term and that this has been well underway and partially achieved for a while now.

No ideology is logical, capitalist ideology is not either. What you need to look at is theory. So, would you say communist theory is not logical?

Seems to be a race to the bottom to me, both on the tech side and the ones adopting the tech. And even if LLMs work badly, firing people because of them does not sound like a winning move for a society anyways.

I think you are projecting. Because I could not care less about any "Hackaton".