Keeping the home directories in another tree has been a thing for a very long time. I was working with Xenix in the early 1990s with a second hard drive, and kept all the home directories on the external hard drive. When I needed to do an OS reinstall, it just a matter of mounting the external file system on the path. Same would apply if you're using NFS or any other network file system.
But this is more necessary with children than it is between gentlemen, as it were. What's changed...?
Amazon is largely populated by employees who got there as their first job out of college and don't know any better. So more children than gentlemen.
What distort? We're running Ubuntu LTS on workstations, and we keep the updates rolling, and have no significant issues. Generally when we want to do feature updates, we don't do in-place upgrades at all, we just build a new image and roll it out. We want complete control over new feature rollouts, including any major new upgrades of key software like LibreOffice.
And honestly, that's pretty much how we were managing Windows prior to beginning the migration. Updates in general are always a risky business, and I've seen upgrades in every OS I've worked with since Windows 3.1 go horribly awry. I've baked Windows systems, Linux systems, BSD systems, and even had to finally give up and reinstall my M1 Mac because the major release upgrade worked about 90%, but there was enough peculiar behavior that it just wasn't worth trying to track down.
In all cases in an enterprise environment, regardless of OS, you don't want feature updates, significant changes to functionality, or installs of major version of updates to software. When it comes to that, you're working in a lab environment, rolling out to a few users to test stability and interoperability, and then pushing them out to all the workstations. This isn't a Linux thing, this is just how an IT department stays sane and doesn't screw up the whole organization's workflow.
There are only a few circumstances I can imagine where LTS support over three or four years would even be desirable, and most of those are pretty niche use cases dealing with specialized equipment or legacy systems. In general, whether it's Linux, BSD, Windows or even MacOS, it's always better to do a full reinstall with the new OS. Heck, by the time Windows Server 2003 went mainstream, only madmen were doing in-place upgrades on domain controllers. The better solution was always to build a new DC and then decommission the old one.
All my worst upgrade disasters in any OS came from in-place upgrades just fucking everything up. At best, it left a lot of old cruft hanging around, at worst it rendered a system almost unusable, and it was usually a bit of both.
The way I'm rolling out Debian and Ubuntu at work these days is just working images. Sometimes there's some funky hardware that requires after I clone an image that needs some intervention, but generally it just works. New images are generated every six months, or when a new release has been tested, rinse and repeat. In the business world you don't give a crap about anything but quick up time, and I have a stack of spares in a closet that get refreshed regularly, and when something blows up, you grab one off the shelf and move on. New OS upgrade, new image.
LLMs are mostly composed of regular old fully connected ANNs, and the remainder, the transformers, are also ANNs.
You're talking theoretical things here, not practical reality.
ANNs certainly can learn the rules of chess, and you can train one to play chess at a level that is generally regarded as superhuman.
No one has ever made an ANN that plays chess at a level that is superhuman. AlphaZero is still primarily a tree searching algorithm with an ANN used to evaluate every node of the tree.
There's also a proof that any 2+ layer ANN of sufficient size can learn any IO function.
To remove the tree search, the ANN would need to be VERY big.
The fact that they don't, or don't do so very well, means that the way they are trained is an inefficient way to learn chess.
LLMs are trained in an inefficient way to learn anything. How many billions of pages did you need to be trained how to read?
That's why it can do this at all. What is impressive is that it can do that even without the sort of specialized training you envision.
How many pages of chess instruction do you think ChatGPT has been trained on? How many pages do you think it would take for it to play a decent game of chess?
I wanted to to GP with "Now ask the Atari chess program to summarize a 10-page PDF".
I pulled out an Atari and did that.
The Atari won because it didn't make any mistakes in the summary.
Usually into offshore accounts..
Usually not into offshore accounts. You are completely wrong here.
When someone buys stock, the money goes to whoever sells the stock. And they use it to buy something else.
No, they use it to make bad loans to sell off at a profit
What percentage of loans made by banks are bad? Show you can think.
You're using a keyboard! How quaint!
