What do they mean what if? It is similarly based to something that has existed for around 4 decades. It of course is at a higher standard of efficiency and able to search through and combine more data but it isn't new. It is just a normal technology and this was why myself and many others were shocked at the initial hype.
The unusual feature of AI now as opposed to the last 4 decades is that it is approaching human intelligence. Assuming that progress continues, exceeding human intelligence will have different economic consequences to being a fair bit worse as was the case mostly.
Okay, so AI isn’t exceptional, but I’m also not exceptional. I run on the same tech base as any old chimpanzee, but at one point our differences in degree turned into one of us remaining “normal” and the other burning the entire planet.
Whether the particular current AI tech is it or not, I have yet to be convinced that the singularity is practically impossible, and as long as things develop in the opposite direction, I get increasingly unnerved.
If you use non-constructive reasoning¹ then you can argue for basically any outcome & even convince yourself that it is inevitable. The basic example is as follows, there is no scientific or physical principle that can prevent the birth of someone much worse than Hitler & therefore if people keep having children one of those children will inevitably be someone who will cause unimaginable death & destruction. My recommendation is to avoid non-constructive inevitability arguments using our current ignorant state of understanding of physical laws as the main premise b/c it's possible to reach any conclusion from that premise & convince yourself that the conclusion is inevitable.
I don't think LLMs are building towards an AI singularity at least.
I also wonder if we can even power an AI singularity. I guess it depends on what the technology is. But it is taking us more energy than really reasonable (in my opinion) just to produce and run frontier LLMs. LLMs are this really weird blend of stunningly powerful, yet with a very clear inadequacy in terms of sentient behaviour.
I think the easiest way to demonstrate that, is that it did not take us consuming the entirety of human textual knowledge, to form a much stronger world model.
The singularity will involve quite a bit more complexity than binary counting, arbitrary words and images, and prediction. These were mirages that will be wiping out both Wall Street and our ecology.
What if this paper actually took things seriously?
A serious paper would start by acknowledging that every previous general-purpose technology required human oversight precisely because it couldn't perceive context, make decisions, or correct errors - capabilities that are AI's core value proposition. It would wrestle with the fundamental tension: if AI remains error-prone enough to need human supervisors, it's not transformative; if it becomes reliable enough to be transformative, those supervisory roles evaporate.
These two Princeton computer scientists, however, just spent 50 pages arguing that AI is like electricity while somehow missing that electricity never learned to fix itself, manage itself, or improve itself - which is literally the entire damn point. They're treating "humans will supervise the machines" as an iron law of economics rather than a temporary bug in the automation process that every profit-maximizing firm is racing to patch.
Sometimes I feel like I'm losing my mind when it's obvious that GPT-5 could do better than Narayanan and Kapoor did in their paper at understanding historical analogies.
I think the "calculator for words" analogy is a good one. It's imperfect since words are inherently ambiguous but then again so is certain forms of digital numbers (floating point anyone?).
Artificial Intelligence is a whole subfield of Computer Science.
Code built of nothing but if/else statements controlling the behavior of game NPCs is AI.
A* search is AI.
NLP is AI.
ML is AI.
Computer vision models are AI.
LLMs are AI.
None of these are AGI, which is what does not yet exist.
One of the big problems underlying the current hype cycle is the overloading of this term, and the hype-men's refusal to clarify that what we have now is not the same type of thing as what Neo fights in the Matrix. (In some cases, because they have genuinely bought into the idea that it is the same thing, and in all cases because they believe they will benefit from other people believing it.)
I think I misinterpreted your comment as not understanding the AI effect, but actually you're just summarizing it kind of concisely and sarcastically?
LLMs are one of the first technologies that makes me think the term "AI effect" needs to be updated to "AGI effect". The effect is still there, but it's undeniable that LLMs are capable of things that seem impossible with classical CS methods, so they get to retain the designation of AI.
One, I doubt your premise ever happens in a meaningfully true and visible way -- but perhaps more important, I'd say you're factually wrong in terms of "what is called AI?"
Among most people, you're thinking of things that were debatably AI, today we have things that are AI (again, not due to any concrete definition, simply due to accepted usage of the term.)
Yeah I can see it being like late 90's and early 2000's for a while. Mostly consulting companies raking in the cash setting up systems for older companies, a ton of flame-out startups, and a few new powerhouses.
Will it change everything? IDK, moving everything self-hosted to the cloud was supposed to make operations a thing of the past, but in a way it just made ops an even bigger industry than it was.
AI being normal technology would be the expected outcome, and it would be nice if it just hurried up and happened so I could stop seeing so much spam around AI actually being something much greater than normal technology
If you read the paper, they make a good case that AI is just a normal technology. They're a bit dissmissive, but they're not alone in that. The AI sector has been all too much hype and far too little substance.
Digital spreadsheets (excel, etc) have done much more to change the world than so-called "artificial intelligence," and on the current trajectory it's difficult to see that changing.
The potentially "explosive" part of AI was that it could be self-improving. Using AI to improve AI, or AI improving itself in an exponential growth until it becomes super-human. This is what the "Singularity" and AI "revolution" is based on.
But in the end, despite saying AI has PhD-level intelligence, the truth is that even AI companies can't get AI to help them improve faster. Anything slower than exponential is proof that their claims aren't true.
Explosions rely on having a lot of energy producing material that can suddenly go off. Even if AI starts self improving it's going to be limited by the amount of energy it can get from the power grid which is kind of maxed out at the moment. It may be exponential growth like weeds growing, ie. gradually and subject to human control, rather than like TNT detonating.
I've come to the conclusion that it is a normal, extremely useful, dramatic improvement over web 1.0. It's going to
1) obsolete search engines powered by marketing and SEO, and give us paid search engines whose selling points are how comprehensive they are, how predictable their queries work (I miss the "grep for the web" they were back when they were useful), and how comprehensive their information sources are.
2) Eliminate the need to call somebody in the Philippines awake in the middle of the night, just for them to read you a script telling you how they can't help you fix the thing they sold you.
3) Allow people to carry local compressed copies of all written knowledge, with 90% fidelity, but with references and access to those paid search engines.
And my favorite part, which is just a footnote I guess, is that everybody can move to a Linux desktop now. The chatbots will tell you how to fix your shit when it breaks, and in a pedagogical way that will gradually give you more control and knowledge of your system than you ever thought you were capable of having. Or you can tell it that you don't care how it works, just fix it. Now's the time to switch.
That's your free business idea for today: LLM Linux support. Train it on everything you can find, tune it to be super-clippy. Charge people $5 a month. The AI that will free you from their AI.
Now we just need to annihilate web 2.0, replace it with peer-to-peer encrypted communications, and we can leave the web to the spammers and the spies.
Well, for starters, it would make The Economist's recent article on "What if AI made the world's economic growth explode?" [1] look like the product of overly credulous suckers for AI hype.
I don't understand, that recent article is a fairly balanced take on what the title literally asks: What if? It examines what if the hype is true, or what if it is partially true. That seems like a good article. You might have a point if the article was, "The hype is true: examining how the world will now change" or something.
There were _so many_ articles in the late 80s and early 90s about how computers were a big waste of money. And again in the late 90s, about how the internet was a waste of money.
We aren't going to know the true consequences of AI until kids that are in high school now enter the work force. The vast majority of people are not capable of completely reordering how they work. Computers did not help Sally Secretary type faster in the 1980s. That doesn't mean they were a waste of money.
AI is probably more of an amplifier for technological change than fire or digital computers; but IDK why we would use a different model for this technology (and teams and coping with change).
> [ "From Comfort Zone to Performance Management" (2009) ] also suggests management styles for each stage (Commanding, Cooperative, Motivational, Directive, Collaborative); and suggests that team performance is described by chained power curves of re-progression through these stages
Transforming, Performing, Reforming, [Adjourning]
Carnal Coping Cycle: Denial, Defense, Discarding, Adaptation, and Internalization
While I feel silly to take seriously something printed in The Economist, I would like to mention that people tend to overestimate the short-term impact of any technology and underestimate its long-term impacts. Maybe AI will follow the same route?
At least within tech, there seem to have been explosive changes and development of new products. While many of these fail, things like agents and other approaches for handling foundation models are only expanding in use cases. Agents themselves are hardly a year old as part of common discourse on AI, though technologists have been building POCs for longer. I've been very impressed with the wave of tools along the lines of Claude Code and friends.
Maybe this will end up relegated to a single field, but from where I'm standing (from within ML / AI), the way in which greenfield projects develop now is fundamentally different as a result of these foundation models. Even if development on these models froze today, MLEs would still likely be prompted to start with feeding something to a LLM, just because it's lightning fast to stand up.
"So a paper published earlier this year by Arvind Narayanan and Sayash Kapoor, two computer scientists at Princeton University, is notable for the unfashionably sober manner in which it treats AI: as "normal technology"."
"Differences about the future of AI are often partly rooted in differing interpretations of evidence about the present. For example, we strongly disagree with the characterization of generative AI adoption as rapid (which reinforces our assumption about the similarity of AI diffusion to past technologies)."
57 comments
[ 3.2 ms ] story [ 91.7 ms ] threadWhether the particular current AI tech is it or not, I have yet to be convinced that the singularity is practically impossible, and as long as things develop in the opposite direction, I get increasingly unnerved.
¹https://gemini.google.com/share/d9b505fef250
I also wonder if we can even power an AI singularity. I guess it depends on what the technology is. But it is taking us more energy than really reasonable (in my opinion) just to produce and run frontier LLMs. LLMs are this really weird blend of stunningly powerful, yet with a very clear inadequacy in terms of sentient behaviour.
I think the easiest way to demonstrate that, is that it did not take us consuming the entirety of human textual knowledge, to form a much stronger world model.
A serious paper would start by acknowledging that every previous general-purpose technology required human oversight precisely because it couldn't perceive context, make decisions, or correct errors - capabilities that are AI's core value proposition. It would wrestle with the fundamental tension: if AI remains error-prone enough to need human supervisors, it's not transformative; if it becomes reliable enough to be transformative, those supervisory roles evaporate.
These two Princeton computer scientists, however, just spent 50 pages arguing that AI is like electricity while somehow missing that electricity never learned to fix itself, manage itself, or improve itself - which is literally the entire damn point. They're treating "humans will supervise the machines" as an iron law of economics rather than a temporary bug in the automation process that every profit-maximizing firm is racing to patch. Sometimes I feel like I'm losing my mind when it's obvious that GPT-5 could do better than Narayanan and Kapoor did in their paper at understanding historical analogies.
I could ask the same thing then. When will you take "AI" seriously and stop attributing the above capabilities to it?
Delusional.
Through this lens it's way more normal
Let's not forget there has been times when if-else statements were considered AI. NLP used to be AI too.
Artificial Intelligence is a whole subfield of Computer Science.
Code built of nothing but if/else statements controlling the behavior of game NPCs is AI.
A* search is AI.
NLP is AI.
ML is AI.
Computer vision models are AI.
LLMs are AI.
None of these are AGI, which is what does not yet exist.
One of the big problems underlying the current hype cycle is the overloading of this term, and the hype-men's refusal to clarify that what we have now is not the same type of thing as what Neo fights in the Matrix. (In some cases, because they have genuinely bought into the idea that it is the same thing, and in all cases because they believe they will benefit from other people believing it.)
LLMs are one of the first technologies that makes me think the term "AI effect" needs to be updated to "AGI effect". The effect is still there, but it's undeniable that LLMs are capable of things that seem impossible with classical CS methods, so they get to retain the designation of AI.
They still are, as far as the marketing department is concerned.
Among most people, you're thinking of things that were debatably AI, today we have things that are AI (again, not due to any concrete definition, simply due to accepted usage of the term.)
Will it change everything? IDK, moving everything self-hosted to the cloud was supposed to make operations a thing of the past, but in a way it just made ops an even bigger industry than it was.
Seems to be the referenced paper?
If so previously discussed here: https://news.ycombinator.com/item?id=43697717
But in the end, despite saying AI has PhD-level intelligence, the truth is that even AI companies can't get AI to help them improve faster. Anything slower than exponential is proof that their claims aren't true.
1) obsolete search engines powered by marketing and SEO, and give us paid search engines whose selling points are how comprehensive they are, how predictable their queries work (I miss the "grep for the web" they were back when they were useful), and how comprehensive their information sources are.
2) Eliminate the need to call somebody in the Philippines awake in the middle of the night, just for them to read you a script telling you how they can't help you fix the thing they sold you.
3) Allow people to carry local compressed copies of all written knowledge, with 90% fidelity, but with references and access to those paid search engines.
And my favorite part, which is just a footnote I guess, is that everybody can move to a Linux desktop now. The chatbots will tell you how to fix your shit when it breaks, and in a pedagogical way that will gradually give you more control and knowledge of your system than you ever thought you were capable of having. Or you can tell it that you don't care how it works, just fix it. Now's the time to switch.
That's your free business idea for today: LLM Linux support. Train it on everything you can find, tune it to be super-clippy. Charge people $5 a month. The AI that will free you from their AI.
Now we just need to annihilate web 2.0, replace it with peer-to-peer encrypted communications, and we can leave the web to the spammers and the spies.
[1] https://www.economist.com/briefing/2025/07/24/what-if-ai-mad...
Neither the OP's URL nor djoldman's archive link allow access to the article!8-((
Computer's Aren't Pulling Their Weight (1991)
There were _so many_ articles in the late 80s and early 90s about how computers were a big waste of money. And again in the late 90s, about how the internet was a waste of money.
We aren't going to know the true consequences of AI until kids that are in high school now enter the work force. The vast majority of people are not capable of completely reordering how they work. Computers did not help Sally Secretary type faster in the 1980s. That doesn't mean they were a waste of money.
Diffusion of innovations: https://en.wikipedia.org/wiki/Diffusion_of_innovations :
> The diffusion of an innovation typically follows an S-shaped curve which often resembles a logistic function.
From https://news.ycombinator.com/item?id=42658336 :
> [ "From Comfort Zone to Performance Management" (2009) ] also suggests management styles for each stage (Commanding, Cooperative, Motivational, Directive, Collaborative); and suggests that team performance is described by chained power curves of re-progression through these stages
Transforming, Performing, Reforming, [Adjourning]
Carnal Coping Cycle: Denial, Defense, Discarding, Adaptation, and Internalization
LLMs may set a record for time between specialized/luxury goods and commodity.
There may be a price floor, but it's not very high.
In My Opinion.
---
Ever think about why restaurants pay someone to wash the dishes?
In my house, I have a machine that does that.
In a restaurant, the machine is too slow, and not compatible with the rest of the system of the restaurant.
Until we hit singularity, AI has to be compatible with the rest of the system.
Maybe this will end up relegated to a single field, but from where I'm standing (from within ML / AI), the way in which greenfield projects develop now is fundamentally different as a result of these foundation models. Even if development on these models froze today, MLEs would still likely be prompted to start with feeding something to a LLM, just because it's lightning fast to stand up.
The paper:
https://thedocs.worldbank.org/en/doc/d6e33a074ac9269e4511e5d...
"Differences about the future of AI are often partly rooted in differing interpretations of evidence about the present. For example, we strongly disagree with the characterization of generative AI adoption as rapid (which reinforces our assumption about the similarity of AI diffusion to past technologies)."