You have a strange notion of what "GNU" is. If you refer to the GNU Hurd kernel, that was indeed released very late, and at a time when it was already clear it had been surpassed by Linux in popularity so far that it would never catch up. But all the many free software packages that were published as GNU projects have been used quite a lot, and still are, and of course software is never "done" because it can always be improved. For the most part, GNU, GPL and FSF has striking success and influence on both commercial and non-commercial software.

You did not really expect those people to put a lot of effort into their Potemkin village?

These project are never about anything technical or practical, they are just vehicles to funnel tax payer's and gullible investor's money into the pockets of some shifty people. Not really different from the frequent occasions where some municipality pays $$$$$$ for some shitty "artwork", made with very little effort by someone who knows someone who...

It is that, plus the activity the money is burned for allows for some PR where politicians can claim they are doing something "innovative". Just like with solar freaking roadways, which everyone having few years of physics education can easily debunk as inefficient, and they never worked anywhere, yet time and again new sponsored projects of the same type pop up somewhere on the planet.

Even when there is a primary/standby setup, still the IPs of both could be part of the DNS response at all times, and if a client connects to a server that does not consider itself as "the primary", that server could tell the client to rather use "the other". That way, an outage of the DNS service would not need to result in an outage of the service - at least as long clients or DNS proxies are willing to deliver a cached resolution response.

So reading this article I felt obliged to run the experiment, what would some LLM "invent" when asked to? Ran the following prompt:

Unsurprisingly, the generated responses read like PR articles from contemporary "startups" that want to do something-something-AI while not having an idea how to technically implement their vision. And of course, the "ideas" sounded awfully similar to what one has seen in fictional literature.

So I guess the patent office will just receive AI-slop in addition to the often dysfunctional "inventions" they already have been receiving from human authors, before.

us-east-1 is more like their traditionally most overcrowded region, that will experience all sorts of scaling or migration-from-legacy-to-newer issues first. To the extent that AWS themselves have recommended to customers that they should preferably deploy their stuff in other regions. Which of course also conveniently earns them $$$ when there is significant data traffic from deployments in us-east-1 to deployments in other regions.

More generally, AWS has an incentive to not let people get the impression that deploying their stuff within one region provides sufficient availability. Because two deployments in two regions earn them more than twice what one deployment in one region puts on the customer's bill. Even if customers are clever enough (and very few are) to not have their second deployment with the same cloud vendor, they can at least charge for the data traffic.

Should not "regulated industries" require more than one deployment in a single region of a single cloud vendor to begin with? I understand that "changing DNS entries" may be a convenient action to take for redirecting users elsewhere when there is an outage of a cheap single deployment, but when "high availability" is asked for, resolving service host names to more than one single address where the service can be reached (all the time, not just during an incident) seems like a necessity to me.

How about taxing per distance and weight? After all, having tires pressing down on roads is what causes them to deteriorate, so both distance and vehicle weight play a big role in how much maintenance roads require. Plus more weight goes along with more micro-plastics being shed by tires into the environment.

There is a spurious word "checks" in above sentence that does not belong there.

The same is true for computer software, and it is not "popular culture" that applauds to AI slop, but tech-incompetent "decision makers", who are all in favor of mediocre code and do not even want "their developers" to be innovative, highly skilled or optimizing diligently. Software enshitification is set to be the top trend for years to come.

Will be interesting to see if earlier research finding that LLMs deteriorate quickly when trained on their own output is now proven correct in reality. It will be impossible for those training LLMs to avoid LLM output in their training material if they want to include contemporary material, which they must to stay attractive for users.

Technically correct, but that might be a bad choice because people experience calculators as something that always provides correct results, and it would require lots of additional sentences to then explain why a correctly (according to the weights) chosen most probable continuation of a text may contain very wrong statements.

If other people use the same prompt and (not randomization enabling) parameters, and have no prior context stored with the LLM, the LLM will generate the same response for them.

Present questions to and corresponding responses from contemporary LLMs to random people on the street, and ask them if they think that generating these responses required thinking. You will find that a vast majority of people will answer "yes" to this, even more so if they are not told the responses were generated by a computer. You and I may know how to spot the hints where LLM generated responses differ from what a human would typically respond with, but that does not matter: If you want to educate people on the risks of over-confidence into LLMs, you cannot convince them by starting with an "AI cannot think" statement, which contradicts their personal experience.

We both certainly agree that this is a huge problem with how LLMs are marketed today. I'm just proposing to not use the claim "AI can't think" as an argument towards those "who don't understand technology", because it will not be a convincing argument to them.

I think there is a big difference here: A deceiving marketing name like "Full Self Driving" evokes a pretty precise expectation of what that thing supposedly does (but does not) in everyone - and it also is pretty easy to precisely define what "Full Self Driving" means (or should mean). The word "think" on the other hand has been used extensively for pretty mundane processes - like a chess software displaying "thinking.." while calculating its next move - and nobody has a precise expectation of what that means, or would feel deceived by finding out the chess software does not emulate a human brain to find its next move.
So better use a slogan like "AI does not think like humans" to headline a critical review of contemporary AI capabilities.