This is from the conclusion of my dissertation from 1997 on using a neural network control system in mineral prcessing.

"Overall, it is now evident that neural networks have been trained haphazardly at best. There is almost no discussion in the text books or with the commercial programs about how to gather a good data set. Usually, the advice is limited to getting all relevant operating modes, or to just get as much as data as possible. This work has pulled together concepts from several sources that together can guarantee to a given confidence level that a neural network can be trained. However, this did not solve most of the operational problems with neural network controllers.

Neural networks cannot run across a change in ores, or more generally, they cannot remain valid when the underlying distribution changes. Since neural networks work on pattern recognition and have no a-priori model from which to work, they have several weaknesses when applied to industrial processes.

They must be trained on all possible combinations of inputs, or they fail in unpredictable ways when they encounter a new condition. In simpler terms, they are not particularly robust."

Unpredictable failure modes (the models are non-linear, you don't know what they are going to do) are not compatible with mineral processing or chemical plants in general.

Current AI researchers threw hardware and electricity at it. The models are much bigger, but still fail the same way.

Well, there are some really bad comedy skits on YouTube that feature Spirit. But, beyond that, no, I can't think of anything tech-related about it. And that includes their aircraft.

Is X number 1 in the world got news? No.
Should it be? No.
Is that relevant as to whether Grok is any good? No.
Is Grok any good? No.

Now that's cleared up...

Honestly, I don't know why Snap is even a thing. Yes, you can roll back to a previous version, but if they'd simply used NILFS2, they could have done exactly the same thing without using anything like as much disk space or complexity.

The article is plausible, LLMs have indeed no semantic understanding or concept of logic.

LLMs can be very effective at spotting inconsistencies, dubious reasoning, and design flaws, but you really really have to work hard at it and do a fair amount of the heavy lifting yourself. LLMs, on their own, are worse engineers than Sinclair Research or Microsoft. And that takes some doing.

Even with significant human input, what they produce is likely to be messy and really requires heavy review before use.

Some of you will remember that I've been setting these AI LLMs a serious engineering challenge. Six months of near-constant iterations and there are still hundreds of issues and it's unclear if the design it has come up with is remotely workable. I've got the project up on github, if anyone wants to amuse themselves.

That, despite the first manned hot air balloon flight having taken place in 1783.

How much polysorbate 80 do you find in nature? Not to mention the old reliable coal-tar dies. And don't forget the trans-saturated fats the government said were good for you up until 2002.

By the way, I have a chemistry degree. If I have to look it up it probably doesn't belong internally.

Without all the electronic sensors, computers and other do-dads that drive up the price, make them brake down and EXPENSIVE to repair!

Agreed, they should be labeling it as Advanced Meat Substitute.

"Vegetable matter highly processed to make it try to resemble meat? Very offputting..."

That's it in a nutshell. The be kind to cows movement ran into the highly processed foods are bad for you movement. Humans have been eating meat since there were humans. Industrially chemically modified foods are a very recent thing. Which one is your body designed to handle?

That version is consistent with what I know about Japanese culture and the USSR's involvement. I'm far more inclined to believe it.

Bombing civilians, even in western nations, has never been effective. In Britain, we don't talk about the Blitz Panic, if the Blitz is referred to at all, it is in the context of unifying the nation's resolve.

Why, then, in a culture that put honour above all else, the emperor above all people, and the military over all mindsets, would bombing a city have any different effect? It's obvious from the firestorms created by the US that it hadn't done, and it astonishes me to this day that the US expected any other result.

Yes, yes they did.

I didn't believe them then either.

I have my doubts about an AI's reliability over the course of 400 years. Even if the hardware holds up, by the time the ship gets there, the AI may moved on to different plans than the ones its creators had in mind :)

That was true back then; is it still true now, a number of years, and many expensive lawsuit-settlements later? (I suppose it's possible that Toyota has learned nothing from the experience, but that doesn't seem like the most likely outcome from a company that generally prides itself on quality and reliability)

Linus should have followed the example of Steve Ballmer, who showed us the proper way to rant