> It could mean the sustainably addressable market is only 5% of the current snake oil hype market.

Yeah, this is true for most if not all hype cycles.

In this case there's an international race towards weaponizing the hype before the competition,
so it's anyone's guess how long the industry stays on nationally-funded life support.

If you want to be fast, and don't mind using the command line, Claude Code is your go-to, but you'll pay through the nose.
If you want it to be free (for now) and want a web interface, go mess around with Vibes.diy, but your apps will only work on their platform unless you do a lotta work to selfhost.
If you don't mind slow, set up something like Ollama and Aider, or Browser-Use, but unless you have a decent GPU the slow here is VERY slow with anything above a small or medium model.

Regardless of the method, LLM-assisted bugs are pervasive and perversely subtle, and thinking through what you want to do and defining it in text for an LLM that barely clears the threshold of "clueless intern" becomes an exercise in frustration as your goal seems close, but never quite close enough to actually achieve without manually re-writing it yourself.
Good luck.

Except because LLMs are technically capable of doing stuff sometimes, marketing would keep hyping the stuff it can sometimes do.

And really to be accurate, the warning label would have to say something like "Unless you're a domain expert who will recognize subtle flaws in reasoning and implementation around this domain, DO NOT ATTEMPT TO USE" and everyone would go "I'm an expert, their marketing people told me so, and I can use this on prod wheeeeeee!" so we're in the same place we are now, with ignorant people using these tools in dangerous ways, and experienced people going "eehhhhhh that's not gonna work quite like you think it is, let's isolate concerns" and being ignored or lambasted for their caution.

A 1st Gen iPod is maybe two decades old, which is about a third of the timeframe being discussed.
Dyson was only founded in 1991, which is 20-31 years after the timeframe being discussed.
Was your math done by LLM, perhaps?

Joking aside...
the engineering for a pocket-sized media player
is an entirely different set of considerations and constraints
than for a piece of yard equipment that spins cutting blades.

Clearly, you speak from a place of obscenely specific competency and luck, if you can maintain not just one but two email servers with "minimal effort".

Unfortunately for your example, you are very much the exception, since you are a domain-specific expert who knows off the top of their head, and has the free time to set up and maintain this sort of thing.

Obviously, most people are not like you, and becoming like you is something people hesitate to attempt.

Have you tried googling it? I only say this as a "please, use their own tools to screw them over" sort of thing, any generic search engine will do if you can't stomach Alphabet's search offering.

When someone is willing to pay a livable wage for picking up litter and cleaning graffiti, you will no longer see litter or graffiti.

But right now, housing, food, and transportation costs are exploding, and wages even for previously excellent careers are flat as organizations fall prey to the hype that a cheap LLM can replace the messy expensive humans.

The problem isn't that there's no work, it's that none of the available work will pay a living wage, because "AI can replace humans" is devaluing work faster than it creates new high paying entry level roles.

Why would entry level wages rise after white collar and blue collar work is replaced by mechanized workers, exactly? Where exactly are the funds to hire a babysitter coming from if the parents are out of work because machine learning hype has tanked their previously lucrative computer programming, manufacturing, or similar careers?

Storing the statistical weightings that allow reproduction of significant portions of the "trained" text is functionally indistinguishable from partial plagurism. I'm curious what you think this problem with the underlying mechanics of math and LLM-type generative output would be, exactly.

How about QA or wasting time for scammers?

Those are both areas where a bit of hallucination could be helpful, rather than harmful...

No, the machine calculates a statistically appropriate behavior for a given set of conditions.

This is coming from ingesting a buttload of training data where humans, among other things, try to exploit one another to prolong their existences.

Or a flaw in our observations of what is clearly a many-layered emergent system. Or a flaw in our understanding of how intelligent-seeming behaviors arise.

Speaking as someone who often travels outside cell data zones and prefers to read things on my own terms, I know I'm in the minority, but the use case Pocket was created to fill does still in fact exist.

However, enshittification means that I'll continue meeting my personal need for this exact functionality with "print to PDF" and other basic tools, because those don't randomly change everything in an attempt to get more of my attention every other Tuesday.

Have you tried LibreWolf?

I remember needing to disable one or two things to make it usable (like re-enabling history and keeping tabs between closing and re-opening the browser) but over all it's become my daily driver.