lich_king

My most straightforward read is that the military simply doesn't want their contractors to have a say in the war doctrine. Raytheon doesn't get to say "you can only bomb the countries we like, and no hitting hospitals or schools". It doesn't necessarily mean the Pentagon wants to bomb hospitals, but they also don't want to lose autonomy.

A less charitable interpretation is that the current doctrine is "China / Russia will build autonomous killbots, so we can't allow a killbot gap".

I'm frankly less concerned about "proper" military uses than I am about the tech bleeding into the sphere of domestic law enforcement, as it inevitably will.

It has nothing to do with predictive policing. I don't understand this example, it has nothing to do with detecting intent. You're looking for evidence of a past misdeed.

False positive and false negative rates are non-zero, as with almost anything, but the tools are pretty good. I encourage you to give them a try. Pangram is a good state-of-the-art choice and you can try it for free. They also publish evals and other data about their approach.

> You can detect if texts from a year ago used AI based on statistical patterns.

I don't understand your argument. The vendors for these detection tools can acquire recent samples from all frontier models just as easily as you can use them to write essays. There's nothing that requires a one-year delay.

> Eliminating any statistically significant difference between a high-quality human-written text and LLM-written text is exactly what the LLMs are being trained for.

I think you're basing this off a fundamental misunderstanding of what these detectors look for. LLMs generate human-like text, but they also generate roughly the same style and content every time for a given prompt, modulo some small amount of nondeterminism. In essence, they are a very predictable human. Ask Gemini or ChatGPT ten times in a row to write an essay about why AI is awesome, and it will probably strike about the same tone every single time, with similar syntax, idioms, etc.

This is what these tools detect: the default output of "hey ChatGPT, write me a school essay about X". This can be evaded with clever prompting to assume a different writing personality, but there's only so much evasion you can do without making the text weird in other ways.

I'm always startled about how HN approaches these topics. When we have a press release from a university about how researchers can detect thoughts via fMRI, we have no issue with the claim. But if a vendor makes a pretty believable claim that there are repetitive statistical patterns in LLM output, it's all of sudden treated the same as palm reading.

The problem isn't that AI detection doesn't work. State of the art in this field is pretty solid. The only issue is that it's probabilistic, so it sometimes fails, and when it does, we have nothing else in situations where you actually want to know if someone put in the work.

So what are you proposing, exactly? That we run a large-scale experiment of "let's see what happens if children don't actually need to learn to do thinking and writing on their own"? The reality is that without some form of compulsion, most kids would rather play video games / scroll through TikTok all day. Or that we move to a vastly more resource-intensive model where every kid is given personalized instruction and watched 1:1?

Eh. In the 1980s and 1990s, the capabilities of the software you could run on your new computer were changing dramatically every two years or so. Completely new types of computer games and productivity software, vastly improved audio and video, more and more real-time functionality.

Nowadays, you really don't get these magical moments when you upgrade, not on the device itself. The upgrade from Windows 10 to Windows 11 was basically just more ads. Games released today look about as good as games released 5-10 years ago. The music-making or photo-editing program you installed back then is still good. Your email works the same as before. In fact, I'm not sure I have a single program on my desktop that feels more capable or more responsive than it did in 2016.

There's some magic with AI, but that's all in the cloud.

I mean this with respect, but if this is your sincere worldview, I'd suggest you step outside the tech news bubble a bit more.

There are some white collar jobs, possibly including many software engineers, that are at risk from gen AI. I guess there's some delicious irony in that, given that not long ago, we've been telling blue-collar workers to deal with economic shifts and learn to code. But short of some LessWrong fanfic about self-assembling nanobot AGI doomsday, there's plenty of other things for humans to do for a living.

I think both things can simultaneously be true. There is a certain inevitability to technological progress. Once you reach a critical mass of collective knowledge, the resulting "thing" will get developed. If not by you, then by someone else.

But also, inevitability is not an argument for complicity. If you personally decide to work on bioweapons, I don't think you can shrug and say "eh, it was going to happen either way". As tech workers, we've really mastered the art of coming up with justifications for what essentially just boils down to "all my friends have gotten rich and now it's my turn".

I've met hundreds of sharp engineers from Facebook, Google, Microsoft, etc. None of them could look me straight in the eye and say "yeah, you know, what we're doing with ad tech is actually good". They just always had an explanation along the lines of "it's not that bad, and besides, if we don't do it, someone else will, and we're the good guys here".

This is the plot of the first derivative of employment. It shows a comparatively small but lasting dip after a massive, prolonged, and unwarranted boom between late 2020 and early 2024 that coincided with the dying breaths of ZIRP.

I have no idea about what's coming, but I wouldn't pay a whole lot of attention to people who are looking at the plots of a highly volatile and cyclic industry that goes through constant boom-and-bust cycles, and are trying to position this as proof that AI is or isn't having an impact.

I'm conflicted about this. On one hand, I think LLMs make it easier to discover explanations that, at least superficially, superficially "click" for you. Sure, they were available before, but maybe in textbooks you needed to pay for (how quaint), or on websites that appeared on the fifth page of search results. Whatever are the externalities of that, in the short term, that part may be a net positive for learners.

On the other hand, learning is doing; if it's not at least a tiny bit hard, it's probably not learning. This is not strictly an LLM problem; it's the same issue I have with YouTube educators. You can watch dazzling visualizations of problems in mathematics or physics, and it feels like you're learning, but you're probably not walking away from that any wiser because you have not flexed any problem-solving muscles and have not built that muscle memory.

I had multiple interactions like that. Someone asked an LLM for an ELI5 and tried to leverage that in a conversation, and... the abstraction they came back feels profound to them, but is useless and wrong.

> I disagree. It's worth asking why some people find brand watches beautiful? Where did they get their sense of aesthetic? Were they born with a congenital preference for RM 16-01 Citron?

There's plenty of art that's celebrated, but also kinda weird and ugly. Is "Vertumnus" by Giuseppe Arcimboldo (1591) also a product of the "brand age"? What about various gargoyles and grotesques on old church buildings?

Some people just like weird art, maybe because they think it reflects their own quirky or rebellious nature. Some of these people have money. I don't see why we need some sort of a cynical theory of a "brand-obsessed culture" at the center of it. How many people in your social circle are obsessed with brands? We might have a brand or two we like, typically because we like the way the products look or work. That's about it.

I know some people who like expensive watches. They talk about the design a lot more than they talk about who made it.

I don't understand the metric they're using. Which is maybe to be expected of an article that looks LLM-written. But they started with ~250 URLs; that's a weirdly small sample. I'm sure there are tens of thousands malicious websites cropping up monthly. And I bet that Safe Browsing flags more than 16% of that?

So how did they narrow it down to that small number? Why these sites specifically?... what's the false positive / negative rate of both approaches? What's even going on?

You're responding out of context. The parent was asking if there are bottlenecks specifically related to scheduling. I explicitly made the point that if there are bottlenecks, they're more likely related to memory.

> I’m willing to bet they will go after the smooth rocks and it’s about rarity, not crystals.

Why? Crystals are pretty, rocks are not. We clearly prefer shiny colorful things to dull beige things, even if shiny things are abundant.

As someone coming from a family of editors and plugged into the publishing world, I think it would be really weird if that was your job. It's not an adversarial relationship. Your job is to pressure-test the arguments and the language, not to ask every time if maybe the person submitting the article didn't really write it, or didn't really interview the person they're claiming to have interviewed.

> A lapse in that non-hypothetically left me responsible, and legally liable, in situations like this.

It didn't. At worst, it exposed the publisher. And the publisher would have the defense that they had the right policies in place and that the misconduct lies with the journalist. Unless it could be shown that you knew about potential issues and still went through with it for political or financial gain, it's a nothingburger.

> But more importantly, why can’t both be at fault?

They can and are. But the parent literally said it's not the author's fault.

I don't think there are any fundamental bottlenecks here. There's more scheduling overhead when you have a hundred processes on a single core than if you have a hundred processes on one hundred cores.

The bottlenecks are pretty much hardware-related - thermal, power, memory and other I/O. Because of this, you presumably never get true "288 core" performance out of this - as in, it's not going to mine Bitcoin 288 as fast as a single core. Instead, you have less context-switching overhead with 288 tasks that need to do stuff intermittently, which is how most hardware ends up being used anyway.

It is not a job of the editor to assume that the author is lying to you.

> This was an institutional error, not an individual reporter's fault.

Ah yes, "the system made me use AI".

I clicked through the author's earlier stories when this first made waves. I obviously had no proof, but I was pretty certain that he's been using LLMs to generate stories for a good while.

When Ars released a statement saying this was an isolated incident, my reaction was "they probably didn't look too hard". I suspect they did, in the end?

> vanishes in soil in just 13 weeks.

Part of the problem with waste management is that we don't really put it in the soil. Your household garbage is mostly biodegradable, but if it ends up buried in a lined pit under tons of other garbage, even paper and orange peels will probably sit there for centuries. I'm not sure it makes much of a difference what kinds or quantities of plastic end up buried in the landfill.

I think the solutions here are more on the supply side than the landfill side. The question there is what are we trying to solve.

Energy use? Most alternative packaging materials are energy-intensive too, so it's less about plastic and more about retail and consumer preferences to have everything individually wrapped and packaged in bags or boxes with colorful graphics, nutrition information, and so on.

Environmental pollution? There, the problem is the plastic that doesn't end up in a landfill. Including our "recycling" shipped overseas.

I'm not sure this analogy holds, for two reasons. First, even in the best case, chain-of-thought transcripts don't reliably tell you what the agent is doing and why it's doing it. Second, if you're dealing with a malicious actor, the transcript may have no relation to the code they're submitting.

The reason you don't have to look at assembly is that the .c file is essentially a 100% reliable and unambiguous spec of how the assembly will look like, and you will be generating the assembly from that .c file as a part of the build process anyway. I don't see how this works here. It adds a lengthy artifact without lessening the need for a code review. It may be useful for investigations in enterprise settings, but in the OSS ecosystem?...

Also, people using AI coding tools to submit patches to open-source projects are weirdly hesitant to disclose that.

I love that you use "fantastic creatures" to describe the world of Jansson, but "warriors" to describe Tolkien. Last time I checked, it had hobbits, dwarves, elves, talking trees... but none of that fantasy nonsense of Moomintrolls, right?

There are some seriously dark themes in there - and unlike in Tolkien, the protagonists are completely helpless when facing them. No epic battle in which magical eagles and a magical bear show up to save the day.

This looks like a vibe-coded promo for the service the parent is associated with, but what cracks me up is that not all this UI clutter is a part of the joke. For example, there are some incessant "chat with AI" bubbles that pop up in the bottom right corner that belong to the platform itself.

Almost any online service has plenty of competition, but it doesn't prevent enshittification once one of them proves that you can squeeze more revenue out of users and get away with it. Netflix charges you the same as five years ago, but you now get ads. You pay for Amazon Prime and get ads. You pay for Spotify, but they now serve you AI music from fake bands to avoid paying royalties to humans. The end game is that all consumer LLMs have ads in the free / cheap tier.

And as other folks are saying, the whole point is that it's a different type of an ad: it's not an annoying pop-up or an unskippable video. It's a subtle recommendation that you don't even notice. High conversion rates, little fatigue... getter than all the cool characters smoking in films a while back.

This appears to be an AI article designed to get clicks. I get that it says something we like, but you're literally upvoting someone for typing in a clickbaity prompt.

I have an opinion about the editorial style of Quanta that I don't think it's popular here (judging by how often they get upvoted), but I think it's a symptom of that.

They cover science, but the template they consistently follow is a vague title that oversells the premise and then an article filled with human-interest details and appeals to implications. This makes it easy for everyone to follow along and have an opinion, but I feel like science is a distant backdrop and never the actual subject.

In this article, what's the one tidbit of scientific knowledge that we gain? Dedekind's and Cantor's work is described only in poetic abstractions ("a wedge he could use to pry open the forbidden gates of infinity"). When the focus is writing a gossip column for eloquent people, precision doesn't matter all that much.

My lived experience with tech companies is that principles are easy when they're free - i.e., when you're telling others what to do, or taking principled stances when a competitor is not breathing down your neck.

So, with all respect, when someone tells me that the people they worked with were well-intentioned and driven by values, I take it with a grain of salt. Been there, said the same things, and then when the company needed to make tough calls, it all fell apart.

However, in this instance, it does seem that Anthropic is walking away from money. I think that, in itself, is a pretty strong signal that you might be right.

> I never heard that

I did, a lot, maybe fifteen years ago. There was a lot of talk about a "3D printing revolution" and being years away from being able to make whatever you want at home. For a while, the "maker" moniker was strongly associated with home manufacturing maximalists.

I still don't get the point the article is making, though. That 3D printer thinking was obviously naive because it underestimated the difficulty of mechanical design and the importance of the economies of scale. Using AI to "write" or "code" is a lot easier than turning a vague idea for a household good into a durable and aesthetic 3D print, so it's apples to oranges.

There are other things that the vibecoding movement is underestimating - when you pay a SaaS vendor, you're usually not paying for code as much as for having a turnkey solution where functionality, security, infrastructure, and user support are someone else's problem. But I think that's pretty much where the parallels end.

It's all this account posts. I don't think the LLM behind it will understand the irony.

This is an inexplicable, AI-written article and the obvious answer is no. There's no performance or complexity overhead to not removing a couple of dead characters. There is a complexity overhead to forking off the list or adding pointless special cases to your code.