This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
I think the problem is that a good portion of the public has these points precisely backwards. They believe in the stupid water and toxin points, while believing that AI is all empty hype that will surely end in a rapture-like bubble pop that will magic everything away and make all technology revert back to 2021 again.
I've had frustrating discussions with people in my life who say that all the data center build out isn't worth it for "a better search engine." I had to make the point that if it is all hype, then they don't need to worry about the environmental effects of data centers, as we'll have a 5-10 year period of build up and usage and then all the infrastructure will be shut down and put on moth balls. The idea that LLMs and generative AI might actually be something new and interesting is not very widespread.
Heck, my D&D group has people who think that Mythos-tier cybercapabilities literally do not exist, and I still have art friends who act as if bad hands are an inevitability for AI art.
The prevalence of really easily-debunked claims is a sign that this will eventually pass. It looks bad now when people are so confidently wrong because it implies they can never be corrected. You can never reason someone out of something they were never reasoned into, but time makes more converts than reason anyways.
It’s fairly easy to make mistakes about what’s real, it happens all the time, but as something becomes more obvious it becomes harder to deny. It’s fine if your friends currently don’t believe in AI capabilities that factually exist. The fact that they do exist will become more and more apparent. The only two options are for them to eventually admit the truth or become crazier and crazier. That might not seem like much, your friends can be wrong forever and it probably won’t affect them much. But it can’t be the basis for a stable long-term political coalition.
People have been denying the validity of the law of supply and demand for centuries (1 2). The market can remain irrational longer than you can remain solvent.
More options
Context Copy link
More options
Context Copy link
Where do you put me, in that I think that data centres themselves are mostly harmless, but also that so-called 'AI' is mostly useless?
Lacking in a bit of imagination and a bit too #nothingeverhappens pilled, but fundamentally reasonable and relying on heuristics that have generally been useful. A year from now, I expect you'll respond to new evidence as it comes in (or enjoy some well-earned gloating time).
The same way that I think of AI boosters as being too credulous, I suppose. I grant that there are some niche uses for the technology that we have seen thus far, but the claims people make about its potential so absurdly exceed anything it has been capable to do thus far, and anything that it looks like it might realistically be able to do in the near-to-medium-term, that I mostly just shake my head at the whole field.
Maybe I'm wrong. When we're all uploaded to the cloud and colonising Proxima Centauri, I'll apologise, I guess.
I suppose the other half of it is a disagreement about the merits of the technology - for instance, considering purely the technology that exists right now (and not guessing anything about the future), I think that it is net a bad thing that LLMs exist.
More options
Context Copy link
More options
Context Copy link
I think being AI skeptical is fine up to a point.
Doubting AI can do things it has been able to do for four months is different from doubting capabilities we haven't seen yet, and might never see. I think there are smart reasons to believe AI might never be significantly more capable than it is (though I don't agree with this position.)
Even from a pure business perspective, I'm open to the idea that AI might be a bubble. But I have forced myself to watch Youtube videos by people confidently predicting that AI will all go away when the bubble pops because it is too expensive to run, and I just feel like they genuinely haven't done the first bit of research on the topic. Like, how hard is it to find out about local models for both text and images?
At the very least, now that these things are out in the wild, there's no way that the open weight LLM and image model genie is ever going back in the bottle. The unit economics of frontier AI may never work, but AI slop is here forever.
Man, if AI levels out at "making all digital public spaces unusable due to a slop overload" I think I'd rather take the paperclips.
More options
Context Copy link
Even at current capabilities AI is transformative. A jack of all trades better than the median human in all of them for 25$ a month that can work 24/7 with information - I am not sure that when it comes to industrialization of physical labor we had seen such jump.
All AI development stops today - the knowledge work and science are transformed. Pun intended.
But is the price $25 a month? I'm seeing stories about how the AI companies are now charging more, and charging on tiers where if you want the best model you have to pay the premium price.
Part of the general doubt is that we were sold the line "cheap miracle" and it's increasingly becoming neither cheap nor a miracle. I do think an AI economy is coming, I don't think it will be "AI will get so smart and recursively self-improving that it can replace nearly all human labour, increase productivity to the moon, prices tumble to the level of 'a groat for a mansion' and we all become our own Musk with our personal AI-powered robot labour that builds anything we want for pennies".
And then you have DS4 flash that is too cheap to meter. I am not what the mark up is on frontier models, but it is probably in the thousands of percent. And more and more it looks like the model is a commodity.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Yeah, I think it's quite possible that Anthropic and or (especially?) OpenAI have over-leveraged themselves and suffer some horrible fate and possibly derail the economy, but I doubt AI will go away, even if its pace of improvement levels off. I think the worst-case scenario is it becomes a niche for hobbyists and cybersecurity types, like Linux or something.
I think @OliveTapenade is wrong inasmuch as coding and cybersecurity is not "niche" but correct inasmuch as software is something like 5% of the US economy and people who are inside of the AI [information] bubble don't realize that not everything is computer.
If 'niche' is too dismissive, would it be better to say 'bounded'? Thus far bots seem to be very good at interacting with data, but they do not interface with the world beyond that. I think I stand by the idea that they're a wordcel technology. Maybe they will put all the coders out of work, but that leaves the overwhelming majority of the economy. I can't help but feel like people who work in data manipulation are over-generalising from their situation to everyone.
This is closer to what I am getting at.
However I am not sure it will stay this way. Using LLMs [and other AI: LLMs may not be the best model to actually accomplish this, and at best they are probably only part of the puzzle] to do object manipulation will likely become a thing. CAD is already real, 3D printing/cutting is already real, throwing LLMs into the mix comes with real risks but also real upsides.
Everyone is very enamored of software, but robots building houses, if possible (and I expect it will be) will probably be where the money is actually at. I think that AI companies have over-focused on using AIs to write software (although, to be fair, it's the lowest hanging fruit) to the detriment of pushing more mundane but vastly more important and, over the long term, remunerative physical labor.
Robot labour will be the game changer, and I don't think we need god-level intelligent AI for that. The problem will be getting robots small enough and manoeuvrable enough that they can replace human labour on construction sites. A robot the size of a crane that can slap up frames faster than a team of builders is great, a robot the size of a crane that you want to be inside that frame wiring up those frames is not.
Yes. I think like 90% of the solution is just going to be "telescoping arms of different sizes."
I think that if robot labor can do 90% of the heavy lifting, if you still need humans to run through and e.g. pull the electrical wire to polish it off, or tape off the windows before painting, etc., it's still going to be a game changer. Obviously "robots do all of the physical labor" is the or at least a dream but I don't think people should let the perfect be the enemy of the good, and if it turns out that there are some niches were human dexterity is just really hard to replace, so be it: we can likely still have the robots welding together the hull or slapping up the frames or what have you.
That's going to be the thing that really changes life for the majority of us, not dreams of "ASI will mean eventually we all have our own robot factories on the moon" style forecasts. And it will be a big hurdle to get over, because if you're going to challenge the construction trades on replacing union labour, you are going to have the unions fighting you. This is more likely to be the AI economy of the near future: first the white collar jobs lost the junior grades because one senior guy with AI can do the work himself, now the blue collar jobs are having bites taken out of them piece by piece. Now maybe it still will be worthwhile to be a plumber, but to be one with his own business, because now they don't take on apprentices anymore, Plumbing Bot is doing the jobs of installation with the boss overseeing work being carried out on several locations at once.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I usually try to say 'LLMs' or 'bots' because I don't think the systems we're currently talking about are 'artificial intelligence'. Unfortunately 'AI' is the term that the companies themselves use and which the media seems to have accepted, but from where I'm standing, calling these systems 'AI' is a kind of marketing gimmick. I think I would reserve the term 'AI' for something that's genuinely intelligent.
In any case, I'm not sure I buy the leap into the material world? We've had CAD for decades. 3D printing is mostly a hobby technology, isn't it? At any rate, the best case scenario here is, what, that you can use an LLM or similar system to help produce models for 3D printing?
You're right, I think, that what would really make a difference is automated construction. But I don't yet see any reason to think that's on the horizon. Manipulating language is one thing, but we still very much struggle to build robots capable of interacting with material reality.
I think it's fair to term "intelligence" as "the capability to solve problems" - which lets us speak meaningfully about objects with even a very small modicum of "intelligence" on a certain axis - things can be "smart" in some senses without that generalizing. For instance, I think I can reasonably say "the LRASM is more intelligent than a Harpoon," because the LRASM is better at solving the problem (correctly identifying a target ship) than the Harpoon is.
LLMs certainly can solve problems, and thus while they are not intelligent and self-aware in the way that humans are I think it is fair to call them artificial intelligence. "Artificial" doesn't just mean "constructed," it also suggests a replacement - just like when you get an artificial leg. When a naval aviator launches an LRASM, its judgment replaces his in the final stages of the combat engagement. I ask an LLM to vibe-code for me, it is replacing my judgment. An artificial leg does not need to be a real leg, or as good as a real leg, to be an artificial leg, it needs to serve the same purpose as a leg.
I do agree, though, that it is helpful to say "LLM" in part because there are plenty of artificial intelligence models that are not large language models, and it is helpful to disambiguate.
Depends on how you define "mostly." Possible most people doing 3D printing are hobbyists. But it's being used for stuff that's pretty serious (buildings, guns, boats).
I think the best case scenario is that you could send your LLM product specifications and it would churn out a product for you. I am skeptical that the best unreleased frontier models could do this for something as simple as a microwave right now. But I could see them being capable of it in the future, particularly if you conceptualize the LLM as being used as a go-between or manager, not the actual design model itself. Walk in to your manufacturing complex one evening and say "I need a light attack aircraft, two-seater, range 500nm unrefueled, payload of no less than 2000 pounds, compatible with our preexisting parts and weapons stockpile and standing technical specifications" and then go to sleep, while the LLMs spend all night working on the design. This could work because the LLM is not trying to create the aircraft de novo. Rather it's building the aircraft in a computer design software that knows the parameters of the facility and the "parts list" and the other relevant technical requirements. It is guarded against hallucination because the tools it is using serve as an external source of truth that will validate its designs. Imagine it as the difference between trying to build a house with a set of LEGO bricks and trying to figure out how to build a house from scratch.
I would guess that an LLM could do something like that in, say, Flyout today. Now, there is a long list of reasons why Flyout is not real life, and thus I want to stress I am talking about a best case scenario that I think it plausible enough with enough development that it warrants consideration. I am not necessarily making a firm prediction.
I think this misunderstands how this words. Today, right now, you can use Blender to create a design to 3D print. And today, right now, LLMs can design 3D objects in Blender. So the LLM-to-physical-world pipeline that matters already exists, it just remains to be seen if it can handle anything more complicated than novelty coat hangers and such. I think this ability is actually much more relevant to scaling up manufacturing than robotic humanoids and thus all of the problems we're still having with things like computer vision and mobility are largely irrelevant.
More options
Context Copy link
This is an inane but weirdly popular talking point that would have been a lot fairer (if lower-stakes and off-puttingly pedantic to normies) to bring up literal decades ago when "AI" was commonly being used to refer to simple single-purpose algorithms in charge of a vaguely anthropomorphic object, like "enemy AI" in a video game. I think it would have been fair to object that those aren't actually AI. But at this point, with the LLMs, people trying to invoke the same argument are dealing with an increasingly clear invisible-unicorn-in-the-garage situation, where the actual underlying belief is that they can't be AI because AI is inherently impossible, but they often know on some level that that's silly because materialism, so they try to launder it through asserting that actually, the essence of real intelligence is the ability to pass whatever specific tests we haven't managed to get the AI up to human level performance at yet.
Like, people know that par-human capability in everything wasn't ever how people defined AI, right? That's closer to a definition of ASI.
I mean, I think that if you had asked people whether enemy AI in a video game is a 'real' or 'true' AI, they would have answered no.
In the past I would have said that what 'AI' implies is something similar to a human intelligence, only in a computer. I think the term implied some sort of self-awareness, consciousness, or sapience. When I say 'AI' in that fuller sense, I think people picture something like HAL 9000, or Isaac Asimov's Robots, or heck, Holly from Red Dwarf. I think what 'AI' implied was something like personhood.
LLMs are not that sort of thing, and I see no particular likelihood of them becoming that sort of thing. So I prefer not to speak as if they are. To do otherwise would be to reify an illusion, I think? ChatGPT or Claude or Gemini are not intelligent, that is, they are not people.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I think they're trying harder than ever, but it just turns out that building general-purpose robots is friggin' hard (sorry, Asimov). Human-level intelligence is a tiny little evolutionary whoopsy that's like a million years old. Language? Maybe 100k years. Programming? Don't make me laugh. But processing sensory data and navigating the real world? Evolution has been selecting for that since the beginning of time. We have, uh, a little catching up to do.
What's funny is that I think LLMs are already at a level where they could handle the high-level organization of a robot's work (albeit perhaps not cost-effectively). e.g. looking at the environment and seeing where the laundry is and where to put it. But they're not very helpful for the dumb low-level motions (what we think of as muscle memory). I believe we're making progress on it, but there hasn't been a giant spike in capability like with LLMs.
I don't actually think you want general-purpose robots, though. What you need is a series of frameworks within which robots can cut, weld, print, fasten, and manipulate objects. These frameworks should go from small to very, very large. The small ones are used to make microwaves and toaster ovens, the medium-sized ones to make cars and airplanes, large (mobile) ones to make houses, and the giant ones to make ships.
General purpose robots are tempting because they don't require replacing preexisting architecture but the truth is that you shouldn't want to retain preexisting architecture. General purpose robots intended to be a 1:1 replacement for humans won't build things much faster than humans (although you can run them on a double shift) and, worse, you'll still need to change up your infrastructure every time you need to make a specific model. If you get a general-purpose framework then changing your production line becomes largely a matter of software redesign.
This is massively significant because it brings down the cost of technological improvement, experimentation, and customization. It also makes industry dual-use again. During World War Two, factories that could make a car or refrigerator were quickly pressed into making tanks and airplanes. We cannot do that today because modern tanks and airplanes are extremely bespoke pieces of equipment. We need manufacturing that is capable of doing that again.
More options
Context Copy link
The technology does generalize surprisingly well, but it's the space with the least good returns on investment and the hardest economic overhead.
Not bad! And what's awkward and expensive now may not be in 3 years.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link