rcoveson

I'm sure you're probably right about video games being used in training and, as a result, the models reflecting video game physics: No air resistance, no static/dynamic friction curve, etc.

However, I don't think that's the problem in the example of the marble bust not toppling. It would have toppled in a video game engine as well; they model gravity and inertia just fine.

I think what's happening is the model isn't being a world model at all, it's being a language model. It's trying to make the "scripted" thing happen, and it probably "feels" like the script should end in the bust on the pedestal.

I doubt you're implying that scientists should "say for certain" even when they aren't certain, or that they shouldn't say anything at all. So I assume you're just saying "misleaders gonna mislead", which is true.

But as long as we're bringing up general social woes, I'll point out that those misleading statements are pretty tame by today's standards. The more popular strategy now is to just lie outright.

I think you're reading the wrong motivation into this and reaching the wrong conclusion.

This behavior, being skeptical and precise, is the very thing that makes scientists more trustworthy than the average Joe with his motivated reasoning.

I'm confident that the people working on this would have made the causal claim if they had been able to collect enough data to support it. They may well be working on gathering data to support the causal claim right now.

I can't remember what show it was (maybe Futurama?), but there's a great bit that lampoons this idea. At a modern art exhibit, somebody leaves something like a banana on an empty pedestal. Later, the attendees are marveling at it, talking about what it means, that sort of thing.

In that is example, there is no artistic intent, but also, the "piece" cannot be readily separated from the other "pieces" in the modern art exhibit that were made with artistic intent.

By your definition of art, people can be "duped" by facts about a piece that don't even accompany many pieces that we think of as "obviously art". They can assume they know the answer, or even a bounding box for the answer like "it was made by a person intentionally", and they can turn out to be wrong, and then... what? It just wasn't art, retroactively?

What if a famous haiku turns out to have been a coded message, 100% instrumental in nature, something like the number and position of enemy forces? "Whoops, guess it wasn't art". What if the best picture I took on a vacation turns out to have been take by a software glitch while my camera sat on my luggage?

We only ever think we know who made something. We do get it wrong. There's a lot of fraud in the art world. Is that likely to get worse or better in the next 100 years, do you think? What are future generations to do, just wonder about whether all their favorite art (that looks like art, feels like art, moves them like art) is "maybe art, maybe not"?

That's silly. There's lots of art with no known creator. Assuming the creator will never be known, it's functionally no different from art that has no creator.

We're already at a point where lots of "art" you look at might have had no human involvement; it could have been generated ad hoc by a web search or in response to the way you hesitated over something else in your feed (okay, maybe not already but certainly in the near future).

But we can't just say that anything that moves you is in a superposition of "art" and "not-art" until sufficient evidence emerges that it was created by a human.

Most people don't experience art through the "art historian" or "true art appreciator" lenses of "who made this and why". They just experience the thing for what it is.

2030's equivalent of Banksy could be a hacker who puts artistic images on major websites. It could then come out, years later, that it was all done by some AI agent. It doesn't matter if that's likely or not, it's obviously possible, and your definition can't just ignore it.

It's a combination of argument from analogy and reductio ad absurdum. It's not a fallacy. You're free to dispute the validity of the analogy, or you can dispute the absurdity, which you did, by confirming that you would indeed love carbon receipts for everything.

No, but figuring out that you're playing a snake-like puzzle game at all in an extremely general input domain and then solving it in the least number of moves definitely feels like evidence of intelligence.

Most of the review is about the art assets, and I doubt the big ones (e.g. the trains themselves) are off-the-rack Unreal assets. An engine like Unreal 5 will cast your assets in harsh relief. Which is to say, if your game's assets look custom and look good in Unreal 5, it does indeed demonstrate effort and skill.

It's a bit of a trick question. Sarcopterygii, the "lobe-finned fishes", are classically represented by the lungfish and the coelacanth and other fishes that are rather distantly related to what we think of as central fishes, like the goldfish.

But the clade also contains all the tetrapods. So valid answers include "Lion" and "Human."

If the LLM answers "lungfish," as they often do, you can follow that up with "what is your favorite animal" and see if it notices the trap: It's stuck answering "lungfish" again or else something outside Sarcopterygii, like a ray-finned fish or a Cnidarian.

> What is anyone expecting by asking an LLM what its favorite anything is?

I imagine that, like me, they're expecting to see what it has to say. You don't think it's interesting which preferences LLMs express and how stable or unstable those preferences are?

There was a time when you could search "the" in Google and the top result would be The Onion. That's obviously a case of either extreme SEO or some kind of expensive deal, but either way it's kind of interesting. But you might say, "what is anyone expecting by Googling the word 'the'?"

And biology is by far the classifier's least favorite topic. It's not even close.

I've had it downgrade to Opus for the following questions:

"How confident are we that English and American Eels both spawn in the Sargasso Sea?"

"Come up with five Zoology questions of increasing difficulty for a trivia game."

"What's your favorite sarcopterygian?"

My wife has some zoology-related preferences in her user instructions, and she had it downgrade to Opus after prompting it with: "plant."

I'm very confused by this comment. The era you're talking about is also the era that Facebook was released and it didn't have a voting system, not even likes/reactions. But that's when the term "social networking" really took off, and it definitely referred to Facebook and not Digg or Reddit or Slashdot, to name another that has a comment voting system.

"Social media" as a term comes even later, to capture Twitter and the social features of YouTube and other stuff like that. But it's all sites where most users are people using real names and real faces, and users generally produce content themselves and follow each other's content.

There's clearly a cluster there and HN/Slashdot/Reddit/Digg are clearly outside it. An umbrella that covers both HN and Facebook is almost meaningless; it's "all websites with user-generated OR user-supplemented content."

I could see this image in my mind before I clicked on it, before even consciously considering what it might be. Is this how an LLM feels on the inside?

No, the worst part of a KVM switch is the video signal switching. You want as few switches in the video signal path as possible and the higher bandwidth you need them to be the more expensive they're going to be. You're already paying for the one in your monitor, so taking advantage of that is the right solution.

IME even high-end KVM switches experience occasional signal interruption or, more often, failure to synchronize at all on output switch.

Do what OP did.

What is this, Humanitarian News?

I don't think that criminal negligence is the most helpful legal tool for incentivizing improved security. It's too hard to prove negligence.

Instead, there should be standard civil penalties for leaking various degrees of PII paid as restitution to the affected individual. Importantly, this must be applied REGARDLESS of "certification" or whether any security practices were "incorrect" or "insufficient". Even if there's a zero-day exploit and you did everything right, you pay. That's the cost of storing people's secrets.

This would make operating services whose whole "thing" is storing a bunch of information about individuals (like Canvas) much more expensive. Good! It's far to cheap to stockpile a ticking time bomb of private info and then walk away paying no damages just because you complied with some out-of-date list of rules or got the stamp of approval from a certification org that's incentivized to give out stamps of approval.

Your acceptance criteria for a last line of defense is pretty strict. Governments break the rules, a lot. If we tried to overthrow every government that broke the rules we'd be in a permanent state of violent revolution.

You don't seem like a violent person; quite the opposite. Most people are like that, myself included. I'm mad at the government. Steaming mad, even. But killing? Not even remotely close.

I get that it's easy to discount low-probability futures as meaningless, but I won't do that either. Maybe the insurance policy that is the 2nd amendment has paid out $0 so far. I don't think that's the case, but for the sake of argument let's say it's so. Even if it were so, history and current events indicate to me that we should keep paying for the policy. We think we've got it bad now, but our guy is an absolute kitten compared to some of the tyrants our species has cooked up.

> That doesn't work with the second amendment though because one side has guns and the other side has guns and tanks and drones and nukes and the ability to control all public communication networks, etc.

I don't want to be too blunt, but this is the "uninformed" I was talking about. The same asymmetry was present in Vietnam, Iraq, and Afghanistan. The ability to level cities is not actually that helpful when the goal is to control the population. Modern revolutions don't involve standing armies that you can kill will tanks.

> outright revolution (at which point the constitution doesn't really matter)

It doesn't matter beyond the point of revolution. It matters a lot that it was in effect before the revolution.

> what can't be argued is that the 2nd amendment has been effective at protecting our rights. Our rights are routinely violated.

I'm not sure if you just don't understand the concept of a last resort or if you actually think that we're at the point of last resort already, in which case my only question is: Do you own a gun yet?

It's as much a fantasy as any other "nuclear option", including the literal nuclear option.

Violent revolutions are a part of our history, and they still happen around the world today. Unless things go very, very poorly in the next few decades, we probably won't see another one in the USA in our lifetimes. We can all admit that that fact makes the 2nd amendment's usefulness feel fantastical.

But on deeper reflection I would hope that we can acknowledge that violent revolution is not an impossibility, it's merely an improbability. And anybody who tries to tell you that hundreds of millions of small arms are inconsequential in a fight is uninformed, to put it lightly.

The fact that the current level of rights abuses (which I would agree is much too high and climbing!) has not lead to a violent revolution is a feature, not a bug.

Who is blaming the students? If 13 year olds were smoking and we blamed poor sports performance in that age group on the smoking, we wouldn't be "blaming" them. We don't model 13 year olds as little islands of free will.

A revelation of a mysterious element of the game which is not revealed in any of its marketing material is a spoiler. The fact that you believe it's a "decent compromise" doesn't enter into it. The proper disclaimer for your comment would be: "Spoilers, but I think these things should be spoiled."

I played the game years ago and did not have this element spoiled, and I thought it was presented at exactly the right time and in the right way. I'd go so far as to say that if somebody is so frustrated by that early mystery (which you're all but guaranteed to understand better and better as you play) that they quit there, then the rest of the game will just be an exercise in misery. It's a puzzle game. The developers put settings in place to cut the flight mechanics out of it so people could just experience it as a puzzle box instead of a flight simulator as well. What they did NOT put in the game is a hint about the thing you're spoiling.

Jury is still out as to how well it works, but the traditional prompt is: "Be fruitful, and multiply"

If I had the thought, in general, that this was a fine thing to do, then yes. Presumably I would do it or permit somebody else to do it and be fired.

I'm happy to say that I would be fired if I did this, thought this, or wrote this comment.

EDIT: Parent used to say "it's common for salespeople to log in to customer environments to show potential customers what the product looks like with actual data in it."

It's even more depressing than that framing would suggest, because we skipped over the decades where cars were just fast, powerful transportation tools and went straight from "mind bicycles" to "mind Teslas" full of cameras, tracking, proprietary software, and subscription fees.

> You see a crow fly into the tongue of a headcrab and die. You now know everything you need to know about this enemy.

But not everything you don't need to know, like it's name. That's a barnacle. But I still love the point your making here. :)

I think that's a little harsh. A lot of the most powerful bits are applicable to any intelligence that we could digitally (ergo casually) instantiate or extinguish.

While it may seem that the origin of those intelligences is more likely to be some kind of reinforcement-learning algorithm trained on diverse datasets instead of a simulation of a human brain, the way we might treat them isn't any less though provoking.

But there's a reason that caches are always sized in powers of two as well, and that same reason is applicable to high-performance ring buffers: Division by powers of two is easy and easy is fast. It's reliably a single cycle, compared to division by arbitrary 32bit integers which can be 8-30 cycles depending on CPU.

Also, there's another benefit downstream of that one: Powers of two work as a schelling point for allocations. Picking powers of two for resizable vectors maximizes "good luck" when you malloc/realloc in most allocators, in part because e.g. a buddy allocator is probably also implemented using power-of-two allocations for the above reason, but also for the plain reason that other users of the same allocator are more likely to have requested power of two allocations. Spontaneous coordination is a benefit all its own. Almost supernatural! :)

Speaking from a place of long-term frustration with Java, some compiler authors just absolutely hate exposing the ability to hint/force optimizations. Never mind that it might improve performance for N-5 and N+5 major releases, it might be meaningless or unhelpful or difficult to maintain in a release ten years from now, so it must not be exposed today.

Just 20 watts, the same amount of electricity that powers 2 LED lightbulbs for 24 hours, one nanosecond, or twelve-thousand years.

Huge oversight by Google. Now they're going to have to invent some other way to indicate that you want to show hidden search results and inodes.