However, if you do that, it is no longer "basic". Now you're back to needing administrators and supervisors to run the fake jobs, having to pay to set up said fake jobs, etc... What, are we going to have them grind in the latest MMO as the job?

Now, I've proposed having a "FedJob" program that does some of what you say - but I try to avoid the broken window fallacy by having them work on "infrastructure" instead. Basically, do work that will make the nation better in the long run. If the economy heats up and hires those people, then the projects can lay fallow instead until the next economic slump.

But the critical part here is that it would be actual wages for actual work.

A big part of the idea behind the UBI is that it does not penalize people for their working status - your proposal would penalize them if they already have a job, just not a well-paying one. If they're busy with a fake job, that means they can't be busy working a real job, looking for a real job, or training for a real job.

Same deal with the housing idea. Odds are it'll be located somewhere that makes getting real employment difficult.

We already have unemployment programs for covering things like the "5 years".

My thought is that with many forms of art being more or less infinitely replicable - anything digital like pictures, music, games, and such can be distributed nearly for free all around the world. Meanwhile, the system of very long copyrights means that anything produced commercially is locked up for longer than we're going to remain alive.
Artistry is a skill as well. I've seen saying that it can take 5k hours to master a new skill. With the "infinitely replicable" thing, it can end up being like some of the scifi stuff which had item duplication - we're even seeing some of it today, in that we're relying on a few experts and NOT training up replacements because there just isn't any work below the expert level.

A few years of supporting "starving artists" to keep them from starving could be the opportunity for them to get the necessary hours in to produce commercially viable work, or be able to release said works into the public domain.
Think Patreon vs Amazon books.

Apparently some moderator lacks a sense of humor. Ah, well, that's how it goes.

I think that depends on how you define "okay life". I tend to reach a lot lower than most people on that.
I still remember getting into a UBI debate with a person who thought that the UBI should pay extra so he could send his kids to a religious school. And not just any religious school, but an expensive one.
He couldn't seem to understand that:
1. UBI is supposed to be universal, not customized to individual people's "needs". Also "basic", IE not much over the minimum needs.
2. That said expensive religious school is a want, not a need. I proposed soliciting donations from other people in his religion if it is considered such a need. IE if the religion demands special schooling, then those in the religion can fund it.
3. My proposed UBI is generally around half that of most other proposals, makes funding it a lot easier.
In general, I'm like "don't want to have roommates, share an apartment to make ends meet? Get a job."

You can't make sensible predictions in a world run by clowns.

It's going to be like in the Simpsons where Homer set up a drinking bird to hit return when various prompts came up so he wouldn't have to actually be at his computer to "work".

Definitely not the way Mazda did it.

With their interface you had physical wheels and buttons, sure, but they were used to move around on a screen. So rather than just a quick glance to tap the screen icon you wanted, you had to watch the screen as you moved over to the selection and "clicked" it.

It's the worst of both worlds.

Honestly, I don't mind touch-based UIs for infotainment as long as they're well-organized and keep the important things in fixed locations, and make the buttons big enough.

Yeah, but how did they fare against the real box-office smash of early 2026: "Melania"?

But it's one louder!

Physics is a lot more powerful than the irrational hallucinations of "audiophiles".

Right, they only want biological neural nets trained on their music. Sure.

You know, this makes me kind of curious. Because any given band will have some position in the latent space, so you can find how close two bands are to each other via the cosine distance between their latent positions.

Open source music models aren't as advanced as the proprietary ones, but I bet you could still repurpose them to do this.

Also, this isn't how AI generation works anyways. You can certainly find bands that a particular song is most similar to (whether human or AI generated music), but AI models don't work by collaging random things together. The sound of a snare drum is based on all snare drums it has ever heard. The sound of a human voice is based on all voices it has ever heard. The particular genre might bias individual aspects toward certain directions (death metal - far more likely to activate circuits associated with male singers, aggressive voices, almost certainly circuits for "growling" tones to the lyrics, etc), but it's not basing even its generation of death metal on just "other death metal songs" (let alone some tiny handful of bands), but rather, everything it has ever heard.

If you're training with a pop song, but the singer briefly growls something out, or briefly the song starts playing death metal-style riffs, that will train the exact same circuits that fire during death metal; neural networks heavily reuse superpositions of states. They're not compartmentalized. But when you're generating with the guidance of "pop", it's very unlikely to trigger the activation of those circuits, whereas if you generate with the guidance of "death metal", it is highly likely to.

Now, a caveat: it's always possible to do overtraining / memorization, and thus learn parts of specific songs, or even whole songs. But that itself comes with caveats. First off, usually your training data volume is vastly larger than your model weights, so you physically can't just memorize it all, and any memorization that does occur (for example, due to a sample being repeatedly duplicated in the dataset) comes at the cost of learning other things. And secondly, as this is a highly undesirable event for trainers (you're wasting compute to get worse results), you monitor loss rates of training data vs. eval data (data that wasn't used in training) to look for signs of memorization (e.g. train loss getting too far below eval loss), and if so, you terminate your training.

Well, they do not want to trigger the crash, do they?

LLMs will always hallucinate. That comes from the math they are using and this problem cannot be fixed.