vintermann

I have had this freeze during a model training session too, when the machine was unattended, so I think it may be GPU-related. I thought for the longest time it was power management related. But maybe that was another freeze bug.

I used to try to train models ages ago (that is, a couple of years), and I never found out how they got good results from top_k in the first place. My problem was that as soon as the model had generated an unlikely token, it was quickly steering towards spaces it had never seen during its training. It found itself in unmapped territory, where anything might happen. With high temperatures, it would then generate noisy garbage, with low temperature it would generate repetitive garbage. We tried to fight this with rollout strategies and backtracking, i.e. if you find yourself in a space where all tokens seem equally likely, you're probably in junk land.

Our tricks didn't work very well. But I didn't manage to keep up well enough to learn what worked.

We are talking about the people who want and might be able to improve bureaucracies (towards some end). I'm asking you: do you assume they're good people? What happens when they're not?

And who's going to do the aligning, if not good people?

    They constantly try to escape
    From the darkness outside and within
    By dreaming of systems so perfect
    That no one will need to be good

If you live in a place which is used only for vacation homes, or worse, only as investment objects, you know the value of owner occupancy.

Non-occupants contribute very little to the actual value of a place - they don't work here, they don't buy here. But they capture much of the value of other people trying to make the place a pleasant place to live, though rising real estate values.

The smart reason you pay people is because you want them to continue producing, not because you want to continue consuming.

I've said this a long time. I also don't want to be "gilding the lily": if someone makes plenty of money making the "content" I like, I would like to redirect my contribution to marginal work, something nice that almost doesn't get made. There should not be "content" multimillionaires, that's a failure of our collective incentive structures.

I have used Linux equally long, Nvidia on my last 3 or 4 systems. They all had problems with very sporadic hard crashes (full system freeze so hard magic sysrq keys don't work, and nothing is visible in the logs. Sometimes with audio looping.) It doesn't happen often, rare enough that I sometimes fooled myself into thinking I'd found and solved the issue. But then 4 months down the road, often with a driver update, suddenly they're back.

Population growth has nothing to do with price increases?

People don't make prices increase, people times money makes prices increase. They're not entirely without money of course, but I don't think it's boat people outbidding you for the homes you want.

Makes you wonder about their "get to the US" lesson learned.

Of course there are degrees of servitude, from light or well-regulated wage payment to outright slavery. (But actual slaves were also often paid wages, something I found pretty surprising and counterintuitive). And if you were a paid servant of someone powerful, that could still carry its own prestige to make up somewhat for the dependence.

To take some examples from genealogy, I had trouble identifying a guy described merely as "servant" in a census as the father of some quite powerful and wealthy people. But it turned out, he was a bailiff, a servant of a fief holder.

I also found out there were laws against not going into servitude in my country. Many, especially young unmarried men, would prefer to work and especially to fish on their own, rather than to work for wages. So the authorities had to resort to actually make it illegal. This wasn't just to be nice to big owners, it was because economies of scale meant lower tax revenues if everyone went fishing on their own little boat.

In general, governments did quite a lot of deliberate effort like that pushing people away from a sustenance economy and into a cash economy where they could produce things rich people (e.g. kings) wanted rather than things they needed themselves.

I guess there's no tradition for publishing "negative results" in mathematics? By that I mean not proof of something negative, but rather, "we tried this thing for ages and couldn't get it to work, but we couldn't prove that it could never work either".

Probably there are understandable reasons for that... But I think "negative science" is really important, and soft results like "we tried that for a long time and it wasn't very fruitful" are actually very important for progress even if they're poorly attested in the written record. I guess in mathematics they come informally from your thesis advisor...

It wasn't just some leftist thinkers, it was most people through history. Working for wages was something slaves did, free men supported themselves with what they could grow or sell, and people routinely preferred what seems to us fairly miserable self-sufficiency over paid servitude.

You don't have to go far into history to find examples, this is stuff you find in regular genealogy.

You could write a JS engine with Zig-like idioms (arena allocation, static initialization)

Could you actually? That seems like a bad fit for a JS engine to me. Predictable memory requirements are great when you can have them, perhaps you can avoid complexity then, but for a JS engine?

It matters what software you're writing. When it's a JavaScript runtime, it's not as if you can get away with preallocating all the memory you'll need like some games used to do.

Is it desecration, though? I'm all in favor of putting remains back once we've learned what we can from them, but I also think opening graves is fine if it's for a good reason (such as archaeology).

I don't approve of putting bodies on display or storing skull collections in cabinets in universities, but before we ship them off to be buried again it's good to take DNA samples!

Thanks to the Gregtech New Horizon folks.

Unless I've got the product genealogy wrong, the rewrite predated Microsoft. The mobile version was also C++ based, and I think it is the distant ancestor to bedrock/windows/console editions.

I remember Greg was actually a pretty controversial modder back in the 1.5 - 1.7 days. Many people did NOT care for the Factorio-like direction he took IC2 in (before Factorio was a thing! Apparently they cited IC2 as an inspiration), and at least once I think he accidentally inflicted that gameplay on modpacks which didn't expect it, removing metal block recipes as I recall. But it's very cool that his mod should get such a legacy.

They've made it easy to fix things too. Just yesterday I ran into a bug in one of the mods they manage, I dreaded setting up deobfuscation and all the stuff necessary to get modding for an ancient forge version going, but nah - they've automated and standardized all that. It was just download, fix and build.

But would you have an equally good experience listening to random beeps and chirps which aren't actually birds? Or from relaxing in the light of a big orange lamp?

It's not really about the artist. It's about the thing that's there. Which you experience, always imperfectly, through your senses. Other people are a wonderful thing that's there, but they aren't the only thing that's enjoyable. Usually the thing that's there which the artist wants you to experience, isn't merely the artist themselves.

Can you wonder at the miracles arising from matrix multiplication? Sure. Especially if it's a good artist trying to show you that miracle. I've enjoyed AI art from the start, but then it's the AI which is the point.

I'm not sure it's the same people being paid for aesthetics as for the art. In streaming, for instance, background music is extremely profitable, and a good example of "just aesthetics". But you rarely see artistic artists crank out a few tracks for the "music for studying" playlists just to pay the bills.

We don't yet know the "load bearing" or "delve" or "not X. Not Y. Just..." Clichés of video AI - at least I don't know. But given it's instruction tuned, we can be confident they will be there, and that the breadth of what it can generate for a given prompt will have extremely low diversity compared to its training data.

Not strictly speaking a quota, but they do say that they do look over their shortlists for diversity (of some unspecified kind) and that they never had to adjust their lists because of it.

That second part is presented as damning evidence that there's nothing like quotas - but all it really means is that they've fully internalized the requirements, whatever they were.

(It reminds me of Nate Silver's recent statement that Disney had almost never interfered in their editorial policy. Nate also thought that was a defense, even though what it really shows is that Disney was willing to interfere but almost never had reason to - Nate being almost perfectly aligned to their interests already.)

I don't know the exact process by which judges are picked, but it's not surprising that they should be so aligned on diversity and what kinds of diversity, just as they will be fairly aligned on ideas of what makes a book enjoyable or good.

Not exactly. I have some voracious readers in my family, but none of them ever tried to publish themselves (despite, in my sister's case, winning an award for a novella written as schoolwork).

The thing is that we want to read the same books. And it's not just about quality. It sucks to be the only one to enjoy a great book, to stand there feeling like you're a Star Trek nerd trying to explain your obsession to a group of rural Siberians.

How could they not? There's no objective way to declare a book great (although there may be some objective ways of declaring a book bad).

Young Adult is a marketing term. It's a really terrible idea to leave classification to marketers.

As it is you've been given a way to respectably say "I'm more mature than you" and you've leapt at the opportunity, it seems.

Aggregating preferences isn't easy. Are you sure you wouldn't get a book which is agreeable to almost anyone but not particularly exciting? This might work for e.g awards for the most important non fiction book, but for fiction, I'd rather try a book someone is excited about.

The ones with power, like Gillibrand, are not "oddballs". They must be seen as exponents of the central idea of the party, which is basically just stable interest on capital. They might once have opposed the allocation of capital into crypto, but if capital is invested in crypto now, then crypto must be defended. Marx's quip about the Tories in England feels appropriate:

"The Tories in England long imagined that they were enthusiastic about monarchy, the church and the beauties of the old English Constitution, until the day of danger wrung from them the confession that they are enthusiastic only about ground rent."

One of my perennial essay ideas, something I've tried to express well (and mostly failed) for decades, is this... Here goes again, I guess.

Immigrants coming to a new culture face hard choices about which customs to keep, and which to let go. If the culture is different enough, they often straight up refuse, and try to hold on to everything they can. This doesn't work. If they try this, from what I've seen their children will probably adopt the worst aspects of both the old culture and the new. What is needed is a conscious choice, about what's important and why, and even more crucially, what's not important.

I tell the story with immigrants because it's easier for people to see it in others. But really, all of us face the same choice, because culture is always changing. We all have to decide what really matters in our culture, what's worth passing on - and what to let go. If we never go against the stream, we end up at the same common low point. But if we always try to go against the stream, we fail.

I think Amish have mostly succeeded in keeping their culture because consciousness about the "why" of their rules is usually (not always!) at front. And they don't reject everything modern, and they're not static in response to a changing world (as this article is a good example of).

I think it's more than reasonable to say that the average minecraft player is probably a child

That is almost certainly not true, especially for the Java version, which this server runs.