Luddites were opposing the owners, not the tech.

Who is this author opposing, pray tell me

I'd say it's still the owners, even if they don't explicitly say or if it's even consciously recognised. I doubt that the tool, put towards broadly positive uses that are considered beneficial and not harmful to individuals or society, would be seen in the same way.

Most fears of AI (in the 2026 sense of the term), and perhaps technology more broadly, are fears of capitalism, ownership, and control, and less about the capabilities of the thing itself.

SEO has happily mutated into LLM training and agentic search optimization, if that's what you're wondering.

"Normal people" (tiresome and unnecessarily reductive meme) do not necessarily care about how it's implemented, but they certainly care about planned obsolescence, which is the target of this law. It just another way to enforce reasonable service life and reduce e-waste. That's the goal of this regulation.

For EVs you need at least a hoist/lifter/crane/other power tool to replace a battery. But sure, there's no actual engineering reason they can't be replaced by the user. Same for the smartwatch - you can replace a battery in most ordinary wristwatches that use them, why not the smart ones? IEMs are usually too small and that's where the engineering limitations might matter. Headphones, no problem.

Those are not mutually exclusive at all, and there were waterproof phones with replaceable batteries (without even needing a screwdriver). This is mostly an excuse.

Then read upthread https://news.ycombinator.com/item?id=47835184

I just don't see why we can't have nice things until proven otherwise (especially considering there is already evidence that this works), rather than have glued-shut devices until proven otherwise (by whom then? Apparently IP and practical experiences aren't enough for you)

I clicked the "parent comment" link all the way to the submission, and opened the submission as well. Nothing mentions IP68. Which "we" is this goalpost coming from?

I had an s5. It was neither waterproof nor even water resistant—the enclosure got too banged up to ensure seals stay sealed.

Current models understand different tokenization variants perfectly, e.g. leading space vs no leading space vs one character per token. It doesn't even affect evals and behchmarks. They're also good at languages that have very flexible word formation (e.g. Slavic) and can easily invent pretty natural non-existent words without being restricted by tokenization. This ability took a bit of a hit with recent RL and code generation optimizations, but this is not related to tokenization.

>None of them realized this duality and have taken one possible interpretation.

I suspect this happens due to mode collapse and has nothing to do with the tokenization. Try this with a base model.

Now make it polar and you almost have a vectorscope...

That's a predictable response, but I think you need to keep up with the times. Modern gaming rigs can do single digit ms click-to-photon latency in hugely complex game engines that have fullscreen shaders, which this thing won't have.

If you're really concerned with the latency, use a modern gaming display and a sub-frame latency retro scaler (if it won't have a builtin one).

Until the large model output variety/mode collapse is solved, tools like this one will be generating monotonous slop. Very few people using it will be specifying each detail of what they want.

There were several points in time (after the SP2 too) when installing WinXP with an active internet connection was nearly impossible, because it would get infected during the installation and shut itself down halfway through it.