To include standing there, menacingly.

Evaporative cooling is a bitch. You have some community about the problems with X, and there’s a range of opinions about how bad X is. Anyone mildly affected won’t post much or stick around. People with intense opinions exaggerate. Whether it’s for comedy or rhetoric, ‘X will be the death of us all!’ chases out even more mild users. Now you have a vicious circle of X haters.

If that’s popular enough to form a meaningful audience, you see careers made, serving that conclusion. Shockingly few of them are grifters. They just posted something honestly critical that the haters really enjoyed, and the likeminded engagement made the author’s brain do the happy chemicals, so now they’re the weekly go-to for obsessively complaining about the evils of X. Still naming any actual problems with X, on par with their original independent criticism… but in the new fire-breathing style that makes even half-true non-issues sound like the worst event in recorded history.

The same can happen for positive attitudes, but the result is less circlejerk, and more… cult. Like that DRSyourGME instance. Or Qanon. When sensible people start toward the exits, that doesn’t mean the party’s over.

Rule: No comparing artificial intelligence/machine learning to simple text prediction algorithms.

That’s an overstep. “Spicy autocorrect” is not a joke exclusive to trolls. LLMs genuinely are simpler than they have any right to be, and it’s ridiculous they work anywhere near this well.

Then again, rigidly defining bad behavior is a poor move anyway, when you’re trying to say “don’t be a tedious asshole.” Tedious assholes will gladly slip around whatever specific problems you name, and bait other people into unwitting violations. The general version of this is enforced civility, i.e. “Rule 1: Be nice! >:(”, and that becomes a duck-blind for infuriating liars. Sometimes “fuck off” is a perfectly reasonable response.

Just write “don’t be a tedious asshole.” Hash out what that means amongst the mod team. Do not be afraid to give people a week-long time-out for things you did not pre-emptively wag a finger about. If they mewl ‘but it didn’t say!,’ tell them, it doesn’t have to. I think everyone is happier when they can trust moderators to make a judgement call on who’s being a dick. And so long as the stakes are temporary, don’t be afraid to get it wrong sometimes.

In case you had any doubts who they’re working for.

2005 post, s/LLM/Google/g.

2035: BASIC supremacy.

Don’t need eight billion parameters to go “But why do you want that?”

I think it’s intended for checking the same bit in multiple bytes. You load the mask instead of the data.

So much 6502 ASM involves turning your brain inside-out… despite being simple, clever, and friendly. Like how you can’t do a strided array sensibly because there’s no address register(s). There is no “next byte.” Naively, you want separate varied data at the same index is separate arrays. Buuut because each read address is absolute, you can do *(&array+1)[n], for free.

What I really miss on NES versus Game Boy is SWAP.

Even on 6502, the BIT command is useless 99% of the time, and AND ~which_bit is the right answer.

Interestingly the Intel MCS-51 ISA did have several bit-addressable bytes. Like a weird zero page.

Nazis getting popular doesn’t make stopping them less necessary.

You’ve never taken a hint in your life.

“… but the trains ran on time!”

What the everloving fuck is wrong with you?

I think they mean whatever’s handling the model. A program into which you feed this inherently restricted format, so it takes advantage of those limitations, in order to run more efficiently.

Like if every number’s magnitude is 1 or 0, you don’t need to do floating-point multiplication.

It’s April. Have you looked outside lately? Reality’s pretty fucking green.

Oblivion is aggressively verdant. The fuck happened?

Brown RPG was the one before that!

Thanks, I stole it.

It’s trinary, and I understand why they instead say “1-bit,” but it still bugs me that they call it “1-bit.”

I’d love to see how low they can push this and still get spooky results. Something with ten million parameters could fit on a Macintosh Classic II - and if it ran at any speed worth calling interactive, it’d undercut a lot of loud complaints about energy use. Training takes a zillion watts. Using the model is like running a video game.

… the foreign-hosted servers, or the models themselves?

There are lingering questions, however, around whether DeepSeek engaged in IP theft to create some of its more competitive models.

They did, just like everybody else.

OpenAI has alleged that the Chinese lab distilled its models, violating OpenAI’s terms of use.

Oh that’s fucking rich.