lennivelkant

For some character sets with a lot of different characters like the Han Unicode representation, that could be cumbersome. Granted, Han might not be a great risk for confusion so you might just whitelist them collectively, but my point is that the approach would have to be more nuanced and complex. Ultimately, humans are complex and so are their languages.

If AI constantly refined its own output, sure, unless it hits a wall eventually or starts spewing bullshit because of some quirk of training. But I doubt it could learn to summarise better without external input, just like a compiler won't produce a more optimised version of itself without human development work.

To clarify, I meant that from the devs' perspective: The effort of individually vetting every single character for possible confusion is immense, and the end result would still be just as western-centric. Imagine having a domain name in Greek where some characters are replaced because they might be confused for Latin characters. Or, conversely, having a few characters replaced by similar Latin ones for an attack, which your solution wouldn't catch.

The result would also still be unreliable even for Westerners. If some other character set you didn't vet also contains similar looking characters, there's a new surface for attack.

To properly close that security gap would be an immense arms race... or you could simply shut down the entire attack vector.

So when you consider the importance of protecting gullible people from insidious attacks and the complexity of trying to allow non-Latin characters without creating openings, the question "How widespread are non-Latin URLs in my target audience and is it critical that they be rendered in their native script?" becomes a calculation of cost and benefit.

It's a shit compromise to deal with the shit fact that some people being assholes ruins good things for the rest of us who aren't.

Yeah but the compilers compile improved versions. Like, if you manually curated the summaries to be even better, then fed it to AI to produce a new summary you also curate... you'll end up with a carefully hand-trained LLM.

Though I guess that would be a lot harder.

From the devs' perspective, the relevant question will be this: How hard is it to map out all the lookalikes, and just how important is it to render foreign domains properly?"

As in, desgined to fail early? I highly doubt that.

Even if it were true, lightbulbs still last longer and are way cheaper. Whether I have to replace them every six years or every five years doesn't matter as much.

You could just google it-

oh wait

Based and Hollywood-pilled

[W]atching atrocities, whoever commits them, is not something most people would want to do.

This is the part about it that upsets me most. Propaganda is bad enough, but propaganda through attempting to force people to watch atrocities?

I personally value being informed, but there is a limit to how much I can take before it becomes second-hand trauma. I can fault nobody for looking away from some of the cruel shit humans can do to each other.

Don't the news feed us too much violence, suffering and death already? Forcing people to watch even more would be heartless without the propaganda angle while propaganda without inflicting psychological trauma would be despicable; combined, they're nothing short of disgusting and despicable.

No bad friends! Focus on the upsides

Eh, I'll worry about that if and when we get to that point. Who knows, maybe they turn out really fuckable.

Sparta couldn't even conquer more than their own backyard. They'd overrun Messenia and enslaved the lot, then spent a few centuries bickering with the rest of Greece, until Persia financed them to claim hegemony. That hegemony lasted 33 years, then they bickered with Thebes for a while, took losses, never quite recovered and eventually got subjugated first by the Macedonians, then by the Romans.