Per the author’s links, he warned that deep learning was hitting a wall in both 2018 and 2022. Now would be a reasonable time to look back and say “whoops, I was wrong about that.” Instead he seems to be doubling down.
Indeed. A mouse that runs through a maze may be right to say that it is constantly hitting a wall, yet it makes constant progress.
An example is citing Mr Sutskever's interview this way:
> in my 2022 “Deep learning is hitting a wall” evaluation of LLMs, which explicitly argued that the Kaplan scaling laws would eventually reach a point of diminishing returns (as Sutskever just did)
which is misleading, since Sutskever said it didn't hit a wall in 2022[0]:
> Up until 2020, from 2012 to 2020, it was the age of research. Now, from 2020 to 2025, it was the age of scaling
The larger point that Mr Marcus makes, though, is that the maze has no exit.
> there are many reasons to doubt that LLMs will ever deliver the rewards that many people expected.
That is something that most scientists disagree with. In fact the ongoing progress on LLMs has already accumulated tremendous utility which may already justify the investment.
a contrarian needs to keep spruiking the point, because if he relents, he loses the core audience that listened to him. That's why it's also the same with those who keep predicting market crashes etc.
Is deep learning approaching a wall? - He doesn't make a concrete prediction, which seems like a hedge to avoid looking silly later. Similarly, I noticed a hedge in this post:
Of course it ain’t over til it’s over. Maybe pure scaling ... will somehow magically yet solve ...
---
But the paper isn't wrong either:
Deep learning thus far is data hungry - yes, absolutely
Deep learning thus far is shallow and has limited capacity for transfer - yes, Sutskeyer is saying that deep learning doesn't generalize as well as humans
Deep learning thus far has no natural way to deal with hierarchical structure - I think this is technically true, but I would also say that a HUMAN can LEARN to use LLMs while taking these limitations into account. It's non-trivial to use them, but they are useful
Deep learning thus far has struggled with open-ended inference - same point as above -- all the limitations are of course open research questions, but it doesn't necessarily mean that scaling was "wrong". (The amount of money does seem crazy though, and if it screws up the US economy, I wouldn't be that surprised)
Deep learning thus far is not sufficiently transparent - absolutely, the scaling has greatly outpaced understanding/interpretability
Deep learning thus far has not been well integrated with prior knowledge - also seems like a valuable research direction
Deep learning thus far cannot inherently distinguish causation from correlation - ditto
Deep learning presumes a largely stable world, in ways that may be problematic - he uses the example of Google Flu Trends ... yes, deep learning cannot predict the future better than humans. That is a key point in the book "AI Snake Oil". I think this relates to the point about generalization -- deep learning is better at regurgitating and remixing the past, rather than generalizing and understanding the future.
Lots of people are saying otherwise, and then when you call them out on their predictions from 2 years ago, they have curiously short memories.
Deep learning thus far works well as an approximation, but its answers often cannot be fully trusted - absolutely, this is the main limitation. You have to verify its answers, and this can be very costly. Deep learning is only useful when verifying say 5 solutions is significantly cheaper than coming up with one yourself.
Deep learning thus far is difficult to engineer with - this is still true, e.g. deep learning failed to solve self-driving ~10 years ago
---
So Marcus is not wrong, and has nothing to apologize for. The scaling enthusiasts were not exactly wrong either, and we'll see what happens to their companies.
It does seem similar to be dot com bubble - when the dust cleared, real value was created. But you can also see that the marketing was very self-serving.
Stuff like "AGI 2027" will come off poorly -- it's an attempt by people with little power to curry favor with powerful people. They are serving as the marketing arm, and oddly not realizing it.
"AI will write all the code" will also come off poorly. Or at least we will realize that software creation != writing code, and software creation is the valuable activity
Several OpenAI people said in 2023 that they were surprised by the acceptance of the public. Because they thought that LLMs were not so impressive.
The public has now caught up with that view. Familiarity breeds contempt, in this case justifiably so.
EDIT: It is interesting that in a submission about Sutskever essentially citing Sutskever is downvoted. You can do it here, but the whole of YouTube will still hate "AI".
What exactly are they doing? I've seen a lot of hype but not much real change. It's like a different way to google for answers and some code generation tossed in, but it's not like LLMs are folding my laundry or mowing my lawn. They seem to be good at putting graphic artists out of work mainly because the public abides the miserable slop produced.
Any fool could have anticipated the eventual result of transformer architecture if pursued to its maximum viable form.
What is impressive is the massive scale of data collection and compute resources rolled out, and the amount of money pouring into all this.
But 10 years ago, spammers were building simple little bots with markov chains to evade filters because their outputs sounded plausibly human enough. Not hard to see how a more advanced version of that could produce more useful outputs.
Any fool could have seen self driving cars coming in 2022. But that didn't happen. And still hasn't happened. But if it did happen, it would be easy to say:
"Any fool could have seen this coming in 2012 if they were paying attention to vision model improvements"
Everyone who lives in the show belt understands that unless a self driving car can navigate icy, snow-covered roads better than humans can, it's a non-starter. And the car can't just "pull over because it's too dangerous" that doesn't work at all.
Interesting to me, during that crazy period when Sutskever ultimately ended up leaving OpenAI, I thought perhaps he had shot himself in the foot to some degree (not that I have any insider information—just playing stupid observer from the outside).
The feeling I have now is that it was a fine decision for him to have made. It made a point at the time, perhaps moral, perhaps political. And now it seems, despite whatever cost there was for him at the time, the "golden years" of OpenAI (and LLM's in general) may have been over anyway.
To be sure, I happen to believe there is a lot of mileage for LLMs even in their current state—a lot of use-cases, integration we have yet to explore. But Sutskever I assume is a researcher and not a plumber—for him the LLM was probably over.
One wonders how long before one of these "break throughs". On one hand, they may come about serendipitously, and serendipity has no schedule. It harkens back to when A.I. itself was always "a decade away". You know, since the 1950's or so.
On the other hand, there are a lot more eyeballs on AI these days than there ever were in Minsky's* day.
(*Hate to even mention the man's name these days.)
I've been conflicted on AI/ML efforts for years. On one hand, the output of locally run inference is astounding. There are plenty of models on HuggingFace that I can run on my Mac Studio and provide real value to me every single work day. On the other hand, while I have the experience to evaluate the output, some of my younger colleagues do not. They are learning, and when I have time to help them, I certainly do, but I wish they just didn't have access to LLMs. LLMs are miracle tools in the right hands. They are dangerous conveniences in the wrong hands.
Wasted money is a totally different topic. If we view LLMs as a business opportunity, they haven't yet paid off. To imply, however, that a massive investment in GPUs is a waste seems flawed. GPUs are massively parallel compute. Were the AI market to collapse, we can imagine these GPUs being sold a severe discounts which would then likely spur some other technological innovation just as the crypto market laid the groundwork for ML/AI. When a resource gets cheap, more people gain access to it and innovation occurs. Things that were previously cost prohibitive become affordable.
So, whether or not we humans achieve AGI or make tons of money off of LLMs is somewhat irrelevant. The investment is creating goods of actual value even if those goods are currently overpriced, and should the currently intended use prove to be poor, a better and more lucrative use will be found in the event of an AI market crash.
Personally, I hope that the AGI effort is successful, and that we can all have a robot house keeper for $30k. I'd gladly trade one of the cars in my household to never do dishes, laundry, lawnmowing, or household repairs again just as I paid a few hundred to never have to vacuum my floors (though I actually still do once a month when I move furniture to vacuum places the Roomba can't go, a humanoid robot could do that for me).
What's the lifecycle length of GPUs? 2-4 years? By the time OpenAIs and Anthropics pivot, many GPUs will be beyond their half-life. I doubt there would be many takers for that infrastructure.
Especially given the humungous scale of infrastructure that the current approach requires. Is there another line of technology that would require remotely as much?
Note, I'm not saying there can't be. It's just that I don't think there are obvious shots at that target.
> On one hand, the output of locally run inference is astounding. There are plenty of models on HuggingFace that I can run on my Mac Studio and provide real value to me every single work day. On the other hand, while I have the experience to evaluate the output, some of my younger colleagues do not. They are learning, and when I have time to help them, I certainly do, but I wish they just didn't have access to LLMs. LLMs are miracle tools in the right hands. They are dangerous conveniences in the wrong hands.
Is weird to me. Surely you recognise just as they don't know what they don't know (which is presumably the problem when it hallucinates), you must also have the same issue, there's just no old greybeard to wish you didn't have access.
Well, I'm the graybeard (literally and metaphorically). I know enough not to blindly trust the LLM, and I know enough to test everything whether written by human or machine. This is not always true of younger professionals.
I don't think it's controversial that these things are valuable but rather the cost to produce use things is up for discussion, and the real problem here. If the price is too high now, then there will be real losses people experience down the line, and real losses have real consequences.
I don't think so about the gpus. It's a sunk cost that won't be repurposed easily--just look at what happened to Nortel. Did all those PBXs get repurposed? Nope--trash. Those data centers are going to eat it hard, that's my prediction. It's not a terrible thing, per se--"we" printed trillions the past few years and those events need a sink to get rid of all the excess liquidity. It's usually a big war, but not always. Last time it was a housing bubble. Everyone was going to get rich on real estate, but not really. It was just an exercise in finding bag holders. That's what this AI/data center situation amounts to as well--companies had billions in cash sitting around doing nothing, might as well spend it. Berkshire has the same problem--hundreds of billions with nowhere to be productively invested. It doesn't sound like a problem but it is.
My humble take on AGI is that we don't understand consciousness so how could we build something conscious except by accident? It seems like an extremely risky and foolish thing to attempt. Luckily, humans will fail at it.
Just because something didn't work out doesn't mean it was a waste, and it isn't particularly clear that the the LLM boom was wasted, or that it is over, or that it isn't working. I can't figure out what people mean when they say "AGI" any more, we appear to be past that. We've got something that seems to be general and seems to be more intelligent than an average human. Apparently AGI means a sort of Einstein-Tolstoy-Jesus hybrid that can ride a unicycle and is far beyond the reach of most people I know.
> We've got something that seems to be general and seems to be more intelligent than an average human.
We've got something that occasionally sounds as if it were more intelligent than an average human. However, if we stick to areas of interest of that average human, they'll beat the machine in reasoning, critical assessment, etc.
And in just about any area, an average human will beat the machine wherever a world model is required, i.e., a generalized understanding of how the world works.
It's not to criticize the usefulness of LLMs. Yet broad statements that an LLM is more intelligent than an average Joe are necessarily misleading.
I like how Simon Wardley assesses how good the most recent models are. He asks them to summarize an article or a book which he's deeply familiar with (his own or someone else's). It's like a test of trust. If he can't trust the summary of the stuff he knows, he can't trust the summary that's foreign to him either.
> Just because something didn't work out doesn't mean it was a waste
Its all about scale.
If you spend $100 on something that didn't work out that money wasn't wasted if you learned something amazing. If you spend $1,000,000,000,000 on something that didn't work out the expectation is that you learn something close to 1,000,000,000x more than the $100 spend. If the value of learning is several orders of magnitude less than the level of investment there is absolutely tremendous waste.
For example: nobody qualifies spending a billion dollars on a failed project as value if your learning only resulted in avoiding future paper cuts.
While it doesn't seem we can agree on a meaning for AGI, I think a lot of people think of it as an intelligent entity that has 100% agency.
Currently we need to direct LLM's from task to task. They don't yet posses the capability of full real world context.
This is why I get confused when people talk about AI replacing jobs. It can replace work, but you still need skilled workers to guide them. To me, this could result in humans being even more valuable to businesses, and result in an even greater demand for labor.
If this is true, individuals need to race to learn how to use AI and use it well.
> Currently we need to direct LLM's from task to task.
Agent-loops that can work from larger scale goals work just fine. We can't letting them run with no oversight, but we certainly also don't need to micro-manage every task. Most days I'll have 3-4 agent-loops running in parallel, executing whole plans, that I only check in on occasionally.
I still need to review their output occasionally, but I certianly don't direct them task to task.
I do agree with you we still need skilled workers to guide them, so I don't think we necessarily disagree all that much, but we're past the point where they need to be micromanaged.
If we can't agree on a definition of AGI, then what good is it to say we have "human-in-the-loop AGI"? The only folks that will agree with you will be using your definition of AGI, which you haven't shared (at least in this posting). So, what is your definition of AGI?
AI capabilities today are jagged and people look at what they want to.
Boosters: it can answer PhD-level questions and it helps me a lot with my software projects.
Detractors: it can't learn to do a task it doesn't already know how to do.
Boosters: But actually it can actually sometimes do things it wouldn't be able to do otherwise if you give it lots of context and instructions.
Detractors: I want it to be able to actually figure out and retain the context itself, without being given detailed instructions every time, and do so reliably.
Boosters: But look, in this specific case it sort of does that.
Detractors: But not in my case.
Boosters: you're just using it wrong. There must be something wrong with your prompting strategy or how you manage context.
AFAICT "AGI" is a placeholder for peoples fears and hopes for massive change caused by AI. The singularity, massive job displacement, et cetera.
None of this is a binary, though. We already have AGI that is superhuman in some ways and subhuman in others. We are already using LLM's to help improve themselves. We already have job displacement.
That continuum is going to continue. AI will become more superhuman in some ways, but likely stay subhuman in others. LLM's will help improve themselves. Job displacement will increase.
Thus the question is whether this rate of change will be fast or slow. Seems mundane, but it's a big deal. Humans can adapt to slow changes, but not so well to fast ones. Thus AGI is a big deal, even if it's a crap stand in for the things people care about.
> Just because something didn't work out doesn't mean it was a waste, and it isn't particularly clear that the the LLM boom was wasted, or that it is over, or that it isn't working
Agreed. Has there been waste? Inarguably. Has the whole thing been a waste? Absolutely not. There are lessons from our past that in an ideal world would have allowed us to navigate this much more efficiently and effectively. However, if we're being honest with ourselves, that's been true of any nascent technology (especially hyped ones) for as long as we've been recording history. The path to success is paved with failure, Hindsight is 20/20, History rhymes and all that.
> I can't figure out what people mean when they say "AGI" any more
We've been asking "What is intelligence" (and/or Sentience) for as long as we've been alive, and still haven't come to a consensus on that. Plenty people will confidently claim they have an answer, which is great, but it's entirely irrelevant if there's not a broad consensus on that definition or a well defined way to verify AI/people/anything against it. Point in case...
> we appear to be past that. We've got something that seems to be general and seems to be more intelligent than an average human
Hard disagree specifically as it regards to Intelligence. They are certainly useful utilities when you use them right, but I digress. What are you basing that on? How can we be sure we're past a goal-post when we don't even know where the goal-post is? For starters, how much is Speed (or latency or IOP/TPSs or however you wish to contextualize it) a function of "intelligence"? For a tangible example of that: If an AI came to a conclusion derived from 100 separate sources, and a human manually went through those same 100 sources and came to the same conclusion, is the AI more intelligent by virtue of completing that task faster? I can absolutely see (and agree with) how that is convenient/useful, but the question specifically is: Does the speed it can provide answers (assuming they're both correct/same) make it smarter or as smart as the human?
How do they rationalize and reason their way through new problems? How do we humans? How important is the reasoning or the "how" of how it arrives at answers to the questions we ask it if the answers are correct? For a tangible example of that: What is happening when you ask an AI to compute the sum of 1 plus 1? What are we doing when we're asking to perform the same task? What about proving it to be correct? More broadly, in the context of AGI/Intelligence, does it matter if the "path of reason" differs if the answers are correct?
What about how confidently it presents those answers (correct or not)? It's well known that us humans are incredibly biased towards confidence. Personally, I might start buying into the hype the day that AI starts telling me "I'm not sure" or "I don't know." We'll get there one day, and until then I'm happy to use it for the utility and convenience it provides while doing my part to make it better and more useful.
> Just because something didn't work out doesn't mean it was a waste
Here i think it's more about opportunity cost.
> I can't figure out what people mean when they say "AGI" any more, we appear to be past that
What i ask of an AGI is to not hallucinate idiotic stuff. I don't care about being bullshitted too much if the bullshit is logic, but when i ask "fix mypy errors using pydantic" and instead of declaring a type for a variable it invent weird algorithms that make no sense and don't work (and the fix would have taken 5 minutes for any average dev).I mean, Claude 4.5 and Codex have replaced my sed/search and replaces, write my sanity tests, write my commit comment, write my migration scripts (and most of my scripts), and make refactor so easy i now do one refactor every month or so, but if it is AGI, i _really_ wonder what people mean by intelligence.
> Also, if anyone wants to know what a real effort to waste a trillion dollars can buy
100% agree. Pleas Altman, Ilya and other, i will hapilly let you use whatever money you want if that money is taken from war profiteers and warmongers.
> Just because something didn't work out doesn't mean it was a waste
One thing to keep in mind, is that most of these people who go around spreading unfounded criticism of LLMs, "Gen-AI" and just generally AI aren't usually very deep into understanding computer science, and even less science itself. In their mind, if someone does an experiment, and it doesn't pan out, they'll assume that means "science itself failed", because they literally don't know how research and science work in practice.
Maybe true in general, but Gary Marcus is an experienced researcher and entrepreneur who’s been writing about AI for literally decades.
I’m quite critical, but I think we have to grant that he has plenty of credentials and understands the technical nature of what he’s critiquing quite well!
Eh, tearing down a straw man is not an impressive argument from you either.
As a counter-point, LLMs still do embarrassing amounts of hallucinations, some of which are quite hilarious. When that is gone and it starts doing web searches -- or it has any mechanisms that mimic actual research when it does not know something -- then the agents will be much closer to whatever most people imagine AGI to be.
Can they, fundamentally, do that? That is, given the current technology.
Architecturally, they don't have a concept of "not knowing." They can say "I don't know," but it simply means that it was the most likely answer based on the training data.
> When that is gone and it starts doing web searches -- or it has any mechanisms that mimic actual research when it does not know something
ChatGPT and Gemini (and maybe others) can already perform and cite web searches, and it vastly improves their performance. ChatGPT is particularly impressive at multi-step web research. I have also witnessed them saying "I can't find the information you want" instead of hallucinating.
It's not perfect yet, but it's definitely climbing human percentiles in terms of reliability.
I think a lot of LLM detractors are still thinking of 2023-era ChatGPT. If everyone tried the most recent pro-level models with all the bells and whistles then I think there would be a lot less disagreement.
I believe in a very practical definition of AGI. AGI is a system capable of RSI. Why? Because it mimics humans. We have some behaviours that are given to us from birth, but the real power of humans is our ability to learn and improve ourselves and the environment around us.
A system capable of self improvement will be sufficient for AGI imo.
Ah - recursive self improvement. I was thinking repetitive strain injury was odd. But that's probably quite a good test although LLMs may be able to improve a bit but still not be very good. An interesting point for me is if all humans went away could the AI/robots keep on without us which would require them to be able to maintain and build power plants, chip fabs and the like. A way to go on that one.
Self improvement doesn’t mean self improvement in any possible direction without any tradeoffs. Genetic algorithms can do everything an LLM can given enough computational resources and training, but being wildly inefficient humanity can’t actually use them to make a chatbot on any even vaguely relevant timeline.
No suprise to see self-congratulations and more "I'm the only person who ever questioned genAI" nonsense as the key parts of this article. What a bore Marcus is.
There was a lot of talk about reaching "peak AI" in early summer of this year.
I guess there is some truth to it. The last big improvement to LLMs was reasoning. It gave the existing models additional capabilities (after some re-training).
We've reached the plateau of tiny incremental updates. Like with smartphones. I sometime still use an iPhone 6s. There is no fundamental difference compared to the most current iPhone generation 10 years later. The 6s is still able to perform most of the tasks you need a smartphone to do. The new ones do it much faster, and everything works better, but the changes are not disrupting at all.
That's like saying a trillion dollars was potentially wasted sending men to the moon. You have to close your eyes to so much obvious progress and dissect your idea beyond recognition to start believing this thesis.
The core argument here, as far as I can discern it, seems to be: A trillion dollars has been spent scaling LLMs in an attempt to create AGI. Since scaling alone looks like it won't produce AGI, that money has been wasted.
This is a frankly bizarre argument. Firstly, it presupposes that _only_ way AI becomes useful is if turns into AGI. But that isn't true: Existing LLMs can do a variety of economically valuable tasks, such as coding, even when not being AGI. Perhaps the economic worth of non-AGI will never equal what it costs to build an operate it, but it seems way too early to make that judgement and declare any non-AGI AI as worthless.
Secondly, even if scaling alone won't reach AGI, that doesn't mean that you can reach AGI _without_ scaling. Even when new and better architectures are developed, it still seems likely that, between two models with an equivalent architecture, the one with more data and compute research will be more powerful. And waiting for better architectures before you try to scale means you will never start. 50 years from now, researchers will have much better architectures. Does that mean we should wait 50 years before trying to scale them? How about 100 years? At what point do you say, we're never going to discover anything better, so now we can try scaling?
companies are already wasting majority fractions of their engineering labor spend on coordination costs and fake work, through that lens i have trouble making an argument that any of this matters. Which is why they are able to do it. I’m reminded of an old essay arguing that the reason Google spends so lavishly is because if they only spent what they needed, they would appear so extraordinarily profitable that the government would intervene.
I think the article makes decent points but I don't agree with the general conclusion here, which is that all of this investment is wasted unless it "reaches AGI." Maybe it isn't necessary for every single dollar we spend on AI/LLM products and services to go exclusively toward the goal of "reaching AGI?" Perhaps it's alright if these dollars instead go to building out useful services and applications based on the LLM technologies we already have.
The author, for whatever reason, views it as a foregone conclusion that every dollar spent in this way is a waste of time and resources, but I wouldn't view any of that as wasted investment at all. It isn't any different from any other trend - by this logic, we may as well view the cloud/SaaS craze of the last decade as a waste of time. After all, the last decade was also fueled by lots of unprofitable companies, speculative investment and so on, and failed to reach any pie-in-the-sky Renaissance-level civilization-altering outcome. Was it all a waste of time?
It's ultimately just another thing industry is doing as demand keeps evolving. There is demand for building the current AI stack out, and demand for improving it. None of it seems wasted.
That’s not what he’s saying, the investors are the ones who have put trillions of dollars into this technology on the premise that it will achieve AGI. People like Sam Altman and Marc Andreesen have been going into podcasts saying AGI is imminent and they’re going to automate every job.
The author did not say every dollar was wasted, he said that LLMs will never meet the current investment returns.
It’s very frustrating to see comments like this attacking strawmans and setting up Motte and Bailey arguments every time there’s AI criticism. “Oh but LLMs are still useful” and “Even if LLMs can’t achieve AGI we’ll figure out something that will eventually.” Yes but that isn’t what Sam and Andreesen and all these VCs have been saying, and now the entire US economy is a big gamble on a technology that doesn’t deliver what they said it would and because the admin is so cozy with VCs we’re probably all going to suffer for the mistakes of a handful of investors who got blinded by dollar signs in their eyes.
Those billions of dollars spent on data centres could instead have hypothetically been spent on researchers with relatively modest computing resources. A few hundred billion dollars will buy you a lot of researchers.
The author quite literally says that the last few years were a "detour" that has wasted a trillion dollars. He explicitly lists building new LLMs, building larger LLMs and scaling LLMs as the problem and source of the waste. So I don't think I am strawmanning his position at all.
It is one thing to say that OpenAI has overpromised on revenues in the short term and another to say that the entire experiment was a waste of time because it hasn't led to AGI, which is quite literally the stance that Marcus has taken in this article.
It's the nature of bubbles to overpromise and underdeliver.
So - do Altman and Andreesen really believe that, or is it just a marketing and investment pitch?
A reminder that OpenAI has its own explicit definition of AGI: "Highly autonomous systems that outperform humans at most economically valuable work."
The MS/OAI agreement quantifies that as "profits of $100bn/yr."
This seems rather stupid to me. If you can get multiple AGIs all generating profits of $100bn a year - roughly half an Apple, or two thirds of a Meta - you no longer have anything recognisable as a pre-AI economy, because most of the white collar population is out of work and no longer earning anything. And most of the blue collar population that depends on white collar earnings is in the same boat.
So you have to ask "Profits from doing what, and selling to whom?"
The Altman pitch seems to be "Promise it first, collect investor cash, and worry about the rest later."
You are making the same strawman attack you are criticising.
The dollars invested are not justified considering TODAYs revenues.
Just like 2 years ago people said NVIDIA stock prices was not justified and a massive bubble considering the revenue from those days. But NVIDIA revenues 10xed, and now the stock price from 2 years ago looks seriously underpriced and a bargain.
You are assuming LLM revenues will remain flat or increase moderately and not explode.
You seem like someone who might be interested in my nuclear fusion startup. Right now all we have is a bucket of water but in five years that bucket is going to power the state of California.
It's not about "every dollar spent" being a waste of time, it's about acknowledging the reality of opportunity cost. Of course, no one in any movement is likely to listen to their detractors, but in this case the pioneers seem to agree.
I think there is broad agreement that new models and architectures are needed, but I don't see it as a waste to also scale the stack that we currently have. That's what Silicon Valley has been doing for the past 50 years - scaling things out while inventing the next set of things - and I don't see this as any different. Maybe current architectures will go the way of the floppy disk, but it wasn't a waste to scale up production of floppy disk drives while they were relevant. And ChatGPT was still released only 3 years ago!
And notably, Marcus has been banging this drum for years. Even this article points back to articles he wrote years ago suggesting deep learning was hitting the wall... With GPT 3....
It's sour grapes because the methods he prefers have not gotten the same attention (hah...) or funding.
He's continuing to push the ludicrous Apple "reasoning paper" that he described as a "knockout blow for LLMs" even though it was nothing of the sort.
With each of his articles, I usually lose more respect for him.
It seems that innovators, researchers, and founders all work at a fast pace, but adoption of new technology, especially LLMs, ends up being done by companies as is convenient. When an open position can go unfilled or a group can scale up without hiring then companies might move forward with a commitment to LLMs.
Even with strong adoption it may take many years for LLMs available now to reach their potential utility in the economy. This should moderate the outlook for future changes, but instead we have a situation where the speculative MIT study that predicted "AI" could perform 12% of the work in the economy is widely considered to not only be accurate, but inevitable in the short term. How much time is needed dramatically changes calculations of potential and what might be considered waste.
Also worth keeping in mind that the Y2K tech bust left behind excess network capacity that ended up being useful later, but the LLM boom can be expected to leave behind poorly considered data centers full of burned out chips which is a very different legacy.
LLMs write all my code now and I just have to review it. Not only has my output 3x'ed at least, I also have zero hesitations now tackling large refactors, or tracking down strange bugs. For example, I recently received a report there was some minor unicode related data corruption in some of our doc in our DBs. It was cosmetic, and low priority, also not a simple task to track down traditionally. But now I just put [llm agent on it, to avoid people accusing me of promoting] on it. It found 3 instances of the corruption across hundreds of documents and fixed them.
I am sure some of you are thinking "that is all slop code". It definitely can be if you don't do your due diligence in review. We have definitely seen a bifurcation of devs who do that, and those who don't, where I am currently working.
But by far the biggest gain is my mental battery is far less drained at the end of the day. No task feels soul crushing anymore.
Personally, coding agents are the greatest invention of my lifetime outside the emergence of the internet.
Did someone say that LLM was the final solution while I wasn’t listening? Am I fantasizing the huge outcry about the terrible danger of AGI? Are people not finding ways to use the current levels of LLM all over the place?
The idea that the trillions are a waste is not exactly fresh. The economic model is still not clear. Alarmists have been shrill and omnipresent. Bankruptcy might be the future of everyone.
But, will we look up one day and say, “Ah never mind” about GPT, Claude, et al? Fat chance. Will no one find a use for a ton of extra compute? I’m pretty sure.
I don’t much dispute any of the facts I skimmed off the article but the conclusion is dumb.
Ironically that MIT study that made the rounds a few months ago ("Study finds 90% of AI pilots fail" you remember), also found that virtually every single worker at every company they studied was using LLMs regularly.
The real takeaway of the study was that workers were using their personal LLM accounts to do work rather than using the AI implementation mess their companies had shat out.
My hypothesis is that general computing frameworks are the next big thing. The powerful GPUs have been mostly black boxes for way too long. A lot of clever people will not want to just throw them away or sell them second-hand and will try to find better ways to utilize them.
I might very well be super wrong. F.ex. NVIDIA is guarding their secrets very well and we have no reason to believe they'll suddenly drop the ball. But it does make me think; IMO a truly general GPU (and open + free) compute has been our area's blind spot for way too long.
Some of the participants may go bust but I very much doubt the highly profitable ones like Google, Apple, Nvidia and Microsoft will. There'll be enough demand for existing LLMs to keep the servers busy. Just writing code which works currently is probably enough to justify a fair chunk of the capacity.
To further your point - I mean honestly if this all ends up being an actual bubble that doesn’t manifest a financial return for the liquidity injectors but instead a massive loss (for the .01% who are in large part putting the cash in), did humanity actually lose?
If it pops it might end up being looked at in the lens of history as one of the largest backdoor/proxy wealth redistributions ever. The capex being spent is in large part going to fund the labor of the unwashed masses, and society is getting the individual productivity and efficiency benefits from the end result models.
I’m particularly thankful for the plethora of open source models I have access to thanks to all this.
I, individually, have realized indisputable substantial benefits from having these tools at my disposal every day. If the whole thing pops, these tools are safely in my possession and I’m better because I have them. Thanks .01%!!
(the reality is I don’t think it will pop in the classic sense, and these days it seems the .01 can never lose. either way, the $1tn can’t be labeled as a waste).
It would be nice if they could burn it on something that didn't require them to buy up the world's supply of DDR5 RAM, and triple prices for everyone else.
Every technological change has been accompanied by an investment boom that resulted in some degree of wasted investment: cars, electricity, mass production of bicycles, it goes on and on.
One point about this is that humans appear unable to understand that this is an efficient outcome because investment booms are a product of uncertainty around the nature of the technological change. You are building something is literally completely new, no-one had any idea what cars consumers would buy so lots of companies started to try and work out that out and that consolidated into competition on cost/scale once that became clear. There is no way to go to the end of that process, there are many people outside the sphere of business who are heavily incentivized to say that we (meaning bureaucrats and regulators) actually know what kind of cars consumers wanted and that all the investment was just a waste.
Another point is that technological change is very politically disruptive. This was a point that wasn't well appreciated...but is hopefully clear with social media. There are a large number of similar situations in history though: printing press, newspapers, etc. Technological change is extremely dangerous if you are a politician or regulator because it results in your power decreasing and, potentially, your job being lost. Again, the incentives are huge.
The other bizarre irony of this is that people will look at an investment boom with no technological change, that was a response to government intervention in financial markets and a malfunctioning supply-side economy...and the response was: all forms of technical innovation are destabilizing, investment booms are very dangerous, etc. When what they mean is corporations with good political connections might lose money.
This is also linked inherently to the view around inflation. The 1870s are regarded as one of the most economically catastrophic periods in economic history by modern interpretations of politics. Let me repeat this in another way: productivity growth was increasing by 8-10%/year, you saw mind-boggling gains from automation (one example is cigarettes, iirc it took one skilled person 10-20 minutes to create a cigarette, a machine was able to produce hundreds in a minute), and conventional macroeconomics views this as bad because...if you can believe it...they argue that price declines lead to declines in investment. Now compare to today: prices continue to rise, investment is (largely) non-existent, shortages in every sector. Would you build a factory in 1870 knowing you could cut prices for output by 95% and produce more? The way we view investment is inextricably linked in economic policy to this point of view, and is why the central banks have spent trillions buying bonds with, in most cases, zero impact on real investment (depending on what you mean, as I say above, private equity and other politically connected incumbents have made out like bandits...through the cycle, the welfare gain from this is likely negative).
You see the result of this all over the Western world: shortages of everything, prices sky-high, and when technological change happens the hysteria around investment being wasteful and disruptive. It would be funny if we didn't already see the issues with this path all around us.
It is not wasted, we need more of this, this ex-post, academic-style reasoning of everything in hindsight gets us nowhere. There is no collateral damage, even in the completely fake Fed-engineered housing bubble, the apparently catastrophic cost was: more houses, and some wealthy people lost their net worth (before some central bankers found out their decisions in 03-04 caused wealthy people to lose money, and quickly set about recapitalising their brokerage accounts with taxpayers money).
Perhaps the scale is unprecedented, or it's always been like this it's just much less concealed these days.
Absolute retards can waste trillions of dollars on stupid ideas, because they're in the in group. Next door someone who's worked their whole life gets evicted because their mortgage is now way more of what they make in salary.
I always love a Marcus hot take, but this one is more infuriating than usual. He’s taking all these prominent engineers saying “we need new techniques to build upon the massive, unexpected success we’ve had”, twisting it into “LLMs were never a success and sucked all along”, and listing them alongside people that no one should be taking seriously — namely, Emily Bender and Ed Zitron.
Of course, he includes enough weasel phrases that you could never nail him down on any particular negative sentiment; LLMs aren’t bad, they just need to be “complemented”. But even if we didn’t have context, the whole thesis of the piece runs completely counter to this — you don’t “waste” a trillion dollars on something that just needs to be complemented!
FWIW, I totally agree with his more mundane philosophical points about the need to finally unify the work of the Scruffies and the Neats. The problem is that he frames it like some rare insight that he and his fellow rebels found, rather than something that was being articulated in depth by one of the fields main leaders 35 years ago[1]. Every one of the tens of thousands of people currently working on “agential” AI knows it too, even if they don’t have the academic background to articulate it.
I look forward to the day when Mr. Marcus can feel like he’s sufficiently won, and thus get back to collaborating with the rest of us… This level of vitriolic, sustained cynicism is just antithetical to the scientific method at this point. It is a social practice, after all!
Yeah, no shit. I’ve been saying this since the day GPT 3 became hyped. I don’t think many with a CS background are buying the “snake oil” of AGI through stochastic parrots.
At some point, even people who hype LLMs will spin their narrative to not look out of touch with reality. Or not more out of touch than is acceptable lol.
If you think scaling is all that matters, you need to learn more about ML.
Read about the the No Free Lunch Theorem. Basically, the reason we need to "scale" so hard is because we're building models that we want to be good at everything. We could build models that are as good at LLMs at a narrow fraction of tasks we ask of them to do, at probably 1/10th the parameters.
> When it comes to machine learning, research has consistently shown, that pretty much the only thing that matters is scaling.
Yes, indeed, that is why all we have done since the 90s is scale up the 'expert systems' we invented ...
That's such an a-historic take it's crazy.
* 1966: failure of machine translation
* 1969: criticism of perceptrons (early, single-layer
artificial neural networks)
* 1971–75: DARPA's frustration with the Speech
Understanding Research program at Carnegie Mellon University
* 1973: large decrease in AI research in the United Kingdom in response to the Lighthill report
* 1973–74: DARPA's cutbacks to academic AI research in general
* 1987: collapse of the LISP machine market
* 1988: cancellation of new spending on AI by the Strategic Computing Initiative
* 1990s: many expert systems were abandoned
* 1990s: end of the Fifth Generation computer project's original goals
Time and time again, we have seen that each academic research begets a degree of progress, improved by the application of hardware and money, but ultimately only a step towards AGI, which ends with a realisation that there's a missing congitive ability that can't be overcome by absurd compute.
Well, expert systems aren’t machine learning, they’re symbolic. You mention perceptrons, but that timeline is proof for the power of scaling, not against — they didn’t start to really work until we built giant computers in the ~90s, and have been revolutionizing the field ever since.
Per the author’s links, he warned that deep learning was hitting a wall in both 2018 and 2022. Now would be a reasonable time to look back and say “whoops, I was wrong about that.” Instead he seems to be doubling down.
The author is a bit of a stopped clock that who has been saying deep learning is hitting a wall for years and I guess one day may be proved right?
He probably makes quite good money as the go to guy for saying AI is rubbish? https://champions-speakers.co.uk/speaker-agent/gary-marcus
Well..... tbf. Each approach has hit a wall. It's just that we change things a bit and move around that wall?
But that's certainly not a nuanced / trustworthy analysis of things unless you're a top tier researcher.
Indeed. A mouse that runs through a maze may be right to say that it is constantly hitting a wall, yet it makes constant progress.
An example is citing Mr Sutskever's interview this way:
> in my 2022 “Deep learning is hitting a wall” evaluation of LLMs, which explicitly argued that the Kaplan scaling laws would eventually reach a point of diminishing returns (as Sutskever just did)
which is misleading, since Sutskever said it didn't hit a wall in 2022[0]:
> Up until 2020, from 2012 to 2020, it was the age of research. Now, from 2020 to 2025, it was the age of scaling
The larger point that Mr Marcus makes, though, is that the maze has no exit.
> there are many reasons to doubt that LLMs will ever deliver the rewards that many people expected.
That is something that most scientists disagree with. In fact the ongoing progress on LLMs has already accumulated tremendous utility which may already justify the investment.
[0]: https://garymarcus.substack.com/p/a-trillion-dollars-is-a-te...
a contrarian needs to keep spruiking the point, because if he relents, he loses the core audience that listened to him. That's why it's also the same with those who keep predicting market crashes etc.
I like how when you click the "key achievements" tab on this site it just says
> 1997 - Professor of Psychology and Neural Science
I thought the point though was that Sutskever is saying it too.
If something hits a wall and then takes a trillion dollars to move forward but it does move forward, I'm not sure I'd say it was just bluster.
Even further back:
> Yet deep learning may well be approaching a wall, much as I anticipated earlier, at beginning of the resurgence (Marcus, 2012)
(From "Deep Learning: A Critical Appraisal")
I read Deep Learning: A Critical Appraisal ? in 2018, and just went back and skimmed it
https://arxiv.org/abs/1801.00631
Here are some of the points
Is deep learning approaching a wall? - He doesn't make a concrete prediction, which seems like a hedge to avoid looking silly later. Similarly, I noticed a hedge in this post:
Of course it ain’t over til it’s over. Maybe pure scaling ... will somehow magically yet solve ...
---
But the paper isn't wrong either:
Deep learning thus far is data hungry - yes, absolutely
Deep learning thus far is shallow and has limited capacity for transfer - yes, Sutskeyer is saying that deep learning doesn't generalize as well as humans
Deep learning thus far has no natural way to deal with hierarchical structure - I think this is technically true, but I would also say that a HUMAN can LEARN to use LLMs while taking these limitations into account. It's non-trivial to use them, but they are useful
Deep learning thus far has struggled with open-ended inference - same point as above -- all the limitations are of course open research questions, but it doesn't necessarily mean that scaling was "wrong". (The amount of money does seem crazy though, and if it screws up the US economy, I wouldn't be that surprised)
Deep learning thus far is not sufficiently transparent - absolutely, the scaling has greatly outpaced understanding/interpretability
Deep learning thus far has not been well integrated with prior knowledge - also seems like a valuable research direction
Deep learning thus far cannot inherently distinguish causation from correlation - ditto
Deep learning presumes a largely stable world, in ways that may be problematic - he uses the example of Google Flu Trends ... yes, deep learning cannot predict the future better than humans. That is a key point in the book "AI Snake Oil". I think this relates to the point about generalization -- deep learning is better at regurgitating and remixing the past, rather than generalizing and understanding the future.
Lots of people are saying otherwise, and then when you call them out on their predictions from 2 years ago, they have curiously short memories.
Deep learning thus far works well as an approximation, but its answers often cannot be fully trusted - absolutely, this is the main limitation. You have to verify its answers, and this can be very costly. Deep learning is only useful when verifying say 5 solutions is significantly cheaper than coming up with one yourself.
Deep learning thus far is difficult to engineer with - this is still true, e.g. deep learning failed to solve self-driving ~10 years ago
---
So Marcus is not wrong, and has nothing to apologize for. The scaling enthusiasts were not exactly wrong either, and we'll see what happens to their companies.
It does seem similar to be dot com bubble - when the dust cleared, real value was created. But you can also see that the marketing was very self-serving.
Stuff like "AGI 2027" will come off poorly -- it's an attempt by people with little power to curry favor with powerful people. They are serving as the marketing arm, and oddly not realizing it.
"AI will write all the code" will also come off poorly. Or at least we will realize that software creation != writing code, and software creation is the valuable activity
Several OpenAI people said in 2023 that they were surprised by the acceptance of the public. Because they thought that LLMs were not so impressive.
The public has now caught up with that view. Familiarity breeds contempt, in this case justifiably so.
EDIT: It is interesting that in a submission about Sutskever essentially citing Sutskever is downvoted. You can do it here, but the whole of YouTube will still hate "AI".
> in this case justifiably so
Oh please. What LLMs are doing now was complete and utter science fiction just 10 years ago (2015).
This.
I’m under the impression that people who are still saying LLMs are unimpressive might just be not using them correctly/effectively.
Or as Primagean says: “skill issue”
Why would the public care what was possible in 2015? They see the results from 2023-2025 and aren't impressed, just like Sutskever.
What exactly are they doing? I've seen a lot of hype but not much real change. It's like a different way to google for answers and some code generation tossed in, but it's not like LLMs are folding my laundry or mowing my lawn. They seem to be good at putting graphic artists out of work mainly because the public abides the miserable slop produced.
Not really.
Any fool could have anticipated the eventual result of transformer architecture if pursued to its maximum viable form.
What is impressive is the massive scale of data collection and compute resources rolled out, and the amount of money pouring into all this.
But 10 years ago, spammers were building simple little bots with markov chains to evade filters because their outputs sounded plausibly human enough. Not hard to see how a more advanced version of that could produce more useful outputs.
Any fool could have seen self driving cars coming in 2022. But that didn't happen. And still hasn't happened. But if it did happen, it would be easy to say:
"Any fool could have seen this coming in 2012 if they were paying attention to vision model improvements"
Hindsight is 20/20.
Everyone who lives in the show belt understands that unless a self driving car can navigate icy, snow-covered roads better than humans can, it's a non-starter. And the car can't just "pull over because it's too dangerous" that doesn't work at all.
I guess I'm worse than a fool then, because I thought it was totally impossible 10 years ago.
> learning was hitting a wall in both 2018 and 2022
He wasn't wrong though.
Interesting to me, during that crazy period when Sutskever ultimately ended up leaving OpenAI, I thought perhaps he had shot himself in the foot to some degree (not that I have any insider information—just playing stupid observer from the outside).
The feeling I have now is that it was a fine decision for him to have made. It made a point at the time, perhaps moral, perhaps political. And now it seems, despite whatever cost there was for him at the time, the "golden years" of OpenAI (and LLM's in general) may have been over anyway.
To be sure, I happen to believe there is a lot of mileage for LLMs even in their current state—a lot of use-cases, integration we have yet to explore. But Sutskever I assume is a researcher and not a plumber—for him the LLM was probably over.
One wonders how long before one of these "break throughs". On one hand, they may come about serendipitously, and serendipity has no schedule. It harkens back to when A.I. itself was always "a decade away". You know, since the 1950's or so.
On the other hand, there are a lot more eyeballs on AI these days than there ever were in Minsky's* day.
(*Hate to even mention the man's name these days.)
I've been conflicted on AI/ML efforts for years. On one hand, the output of locally run inference is astounding. There are plenty of models on HuggingFace that I can run on my Mac Studio and provide real value to me every single work day. On the other hand, while I have the experience to evaluate the output, some of my younger colleagues do not. They are learning, and when I have time to help them, I certainly do, but I wish they just didn't have access to LLMs. LLMs are miracle tools in the right hands. They are dangerous conveniences in the wrong hands.
Wasted money is a totally different topic. If we view LLMs as a business opportunity, they haven't yet paid off. To imply, however, that a massive investment in GPUs is a waste seems flawed. GPUs are massively parallel compute. Were the AI market to collapse, we can imagine these GPUs being sold a severe discounts which would then likely spur some other technological innovation just as the crypto market laid the groundwork for ML/AI. When a resource gets cheap, more people gain access to it and innovation occurs. Things that were previously cost prohibitive become affordable.
So, whether or not we humans achieve AGI or make tons of money off of LLMs is somewhat irrelevant. The investment is creating goods of actual value even if those goods are currently overpriced, and should the currently intended use prove to be poor, a better and more lucrative use will be found in the event of an AI market crash.
Personally, I hope that the AGI effort is successful, and that we can all have a robot house keeper for $30k. I'd gladly trade one of the cars in my household to never do dishes, laundry, lawnmowing, or household repairs again just as I paid a few hundred to never have to vacuum my floors (though I actually still do once a month when I move furniture to vacuum places the Roomba can't go, a humanoid robot could do that for me).
What's the lifecycle length of GPUs? 2-4 years? By the time OpenAIs and Anthropics pivot, many GPUs will be beyond their half-life. I doubt there would be many takers for that infrastructure.
Especially given the humungous scale of infrastructure that the current approach requires. Is there another line of technology that would require remotely as much?
Note, I'm not saying there can't be. It's just that I don't think there are obvious shots at that target.
> On one hand, the output of locally run inference is astounding. There are plenty of models on HuggingFace that I can run on my Mac Studio and provide real value to me every single work day. On the other hand, while I have the experience to evaluate the output, some of my younger colleagues do not. They are learning, and when I have time to help them, I certainly do, but I wish they just didn't have access to LLMs. LLMs are miracle tools in the right hands. They are dangerous conveniences in the wrong hands.
Is weird to me. Surely you recognise just as they don't know what they don't know (which is presumably the problem when it hallucinates), you must also have the same issue, there's just no old greybeard to wish you didn't have access.
Well, I'm the graybeard (literally and metaphorically). I know enough not to blindly trust the LLM, and I know enough to test everything whether written by human or machine. This is not always true of younger professionals.
There's a big difference between:
"creating goods of actual value"
and
"creating goods of actual value for any price"
I don't think it's controversial that these things are valuable but rather the cost to produce use things is up for discussion, and the real problem here. If the price is too high now, then there will be real losses people experience down the line, and real losses have real consequences.
I don't think so about the gpus. It's a sunk cost that won't be repurposed easily--just look at what happened to Nortel. Did all those PBXs get repurposed? Nope--trash. Those data centers are going to eat it hard, that's my prediction. It's not a terrible thing, per se--"we" printed trillions the past few years and those events need a sink to get rid of all the excess liquidity. It's usually a big war, but not always. Last time it was a housing bubble. Everyone was going to get rich on real estate, but not really. It was just an exercise in finding bag holders. That's what this AI/data center situation amounts to as well--companies had billions in cash sitting around doing nothing, might as well spend it. Berkshire has the same problem--hundreds of billions with nowhere to be productively invested. It doesn't sound like a problem but it is.
My humble take on AGI is that we don't understand consciousness so how could we build something conscious except by accident? It seems like an extremely risky and foolish thing to attempt. Luckily, humans will fail at it.
Just because something didn't work out doesn't mean it was a waste, and it isn't particularly clear that the the LLM boom was wasted, or that it is over, or that it isn't working. I can't figure out what people mean when they say "AGI" any more, we appear to be past that. We've got something that seems to be general and seems to be more intelligent than an average human. Apparently AGI means a sort of Einstein-Tolstoy-Jesus hybrid that can ride a unicycle and is far beyond the reach of most people I know.
Also, if anyone wants to know what a real effort to waste a trillion dollars can buy ... https://costsofwar.watson.brown.edu/
> We've got something that seems to be general and seems to be more intelligent than an average human.
We've got something that occasionally sounds as if it were more intelligent than an average human. However, if we stick to areas of interest of that average human, they'll beat the machine in reasoning, critical assessment, etc.
And in just about any area, an average human will beat the machine wherever a world model is required, i.e., a generalized understanding of how the world works.
It's not to criticize the usefulness of LLMs. Yet broad statements that an LLM is more intelligent than an average Joe are necessarily misleading.
I like how Simon Wardley assesses how good the most recent models are. He asks them to summarize an article or a book which he's deeply familiar with (his own or someone else's). It's like a test of trust. If he can't trust the summary of the stuff he knows, he can't trust the summary that's foreign to him either.
> Just because something didn't work out doesn't mean it was a waste
Its all about scale.
If you spend $100 on something that didn't work out that money wasn't wasted if you learned something amazing. If you spend $1,000,000,000,000 on something that didn't work out the expectation is that you learn something close to 1,000,000,000x more than the $100 spend. If the value of learning is several orders of magnitude less than the level of investment there is absolutely tremendous waste.
For example: nobody qualifies spending a billion dollars on a failed project as value if your learning only resulted in avoiding future paper cuts.
It's not waste, it's a way to get rid of excess liquidity caused by massive money printing operations.
We currently have human-in-the-loop AGI.
While it doesn't seem we can agree on a meaning for AGI, I think a lot of people think of it as an intelligent entity that has 100% agency.
Currently we need to direct LLM's from task to task. They don't yet posses the capability of full real world context.
This is why I get confused when people talk about AI replacing jobs. It can replace work, but you still need skilled workers to guide them. To me, this could result in humans being even more valuable to businesses, and result in an even greater demand for labor.
If this is true, individuals need to race to learn how to use AI and use it well.
> Currently we need to direct LLM's from task to task.
Agent-loops that can work from larger scale goals work just fine. We can't letting them run with no oversight, but we certainly also don't need to micro-manage every task. Most days I'll have 3-4 agent-loops running in parallel, executing whole plans, that I only check in on occasionally.
I still need to review their output occasionally, but I certianly don't direct them task to task.
I do agree with you we still need skilled workers to guide them, so I don't think we necessarily disagree all that much, but we're past the point where they need to be micromanaged.
If we can't agree on a definition of AGI, then what good is it to say we have "human-in-the-loop AGI"? The only folks that will agree with you will be using your definition of AGI, which you haven't shared (at least in this posting). So, what is your definition of AGI?
AI capabilities today are jagged and people look at what they want to.
Boosters: it can answer PhD-level questions and it helps me a lot with my software projects.
Detractors: it can't learn to do a task it doesn't already know how to do.
Boosters: But actually it can actually sometimes do things it wouldn't be able to do otherwise if you give it lots of context and instructions.
Detractors: I want it to be able to actually figure out and retain the context itself, without being given detailed instructions every time, and do so reliably.
Boosters: But look, in this specific case it sort of does that.
Detractors: But not in my case.
Boosters: you're just using it wrong. There must be something wrong with your prompting strategy or how you manage context.
etc etc etc...
AFAICT "AGI" is a placeholder for peoples fears and hopes for massive change caused by AI. The singularity, massive job displacement, et cetera.
None of this is a binary, though. We already have AGI that is superhuman in some ways and subhuman in others. We are already using LLM's to help improve themselves. We already have job displacement.
That continuum is going to continue. AI will become more superhuman in some ways, but likely stay subhuman in others. LLM's will help improve themselves. Job displacement will increase.
Thus the question is whether this rate of change will be fast or slow. Seems mundane, but it's a big deal. Humans can adapt to slow changes, but not so well to fast ones. Thus AGI is a big deal, even if it's a crap stand in for the things people care about.
I think when people say "AGI" they might mean synthesis [1]. I'm not sure I have seen that yet in LLMs. Someone correct me if I'm wrong.
[1] https://en.wikipedia.org/wiki/Bloom's_taxonomy
> Just because something didn't work out doesn't mean it was a waste, and it isn't particularly clear that the the LLM boom was wasted, or that it is over, or that it isn't working
Agreed. Has there been waste? Inarguably. Has the whole thing been a waste? Absolutely not. There are lessons from our past that in an ideal world would have allowed us to navigate this much more efficiently and effectively. However, if we're being honest with ourselves, that's been true of any nascent technology (especially hyped ones) for as long as we've been recording history. The path to success is paved with failure, Hindsight is 20/20, History rhymes and all that.
> I can't figure out what people mean when they say "AGI" any more
We've been asking "What is intelligence" (and/or Sentience) for as long as we've been alive, and still haven't come to a consensus on that. Plenty people will confidently claim they have an answer, which is great, but it's entirely irrelevant if there's not a broad consensus on that definition or a well defined way to verify AI/people/anything against it. Point in case...
> we appear to be past that. We've got something that seems to be general and seems to be more intelligent than an average human
Hard disagree specifically as it regards to Intelligence. They are certainly useful utilities when you use them right, but I digress. What are you basing that on? How can we be sure we're past a goal-post when we don't even know where the goal-post is? For starters, how much is Speed (or latency or IOP/TPSs or however you wish to contextualize it) a function of "intelligence"? For a tangible example of that: If an AI came to a conclusion derived from 100 separate sources, and a human manually went through those same 100 sources and came to the same conclusion, is the AI more intelligent by virtue of completing that task faster? I can absolutely see (and agree with) how that is convenient/useful, but the question specifically is: Does the speed it can provide answers (assuming they're both correct/same) make it smarter or as smart as the human?
How do they rationalize and reason their way through new problems? How do we humans? How important is the reasoning or the "how" of how it arrives at answers to the questions we ask it if the answers are correct? For a tangible example of that: What is happening when you ask an AI to compute the sum of 1 plus 1? What are we doing when we're asking to perform the same task? What about proving it to be correct? More broadly, in the context of AGI/Intelligence, does it matter if the "path of reason" differs if the answers are correct?
What about how confidently it presents those answers (correct or not)? It's well known that us humans are incredibly biased towards confidence. Personally, I might start buying into the hype the day that AI starts telling me "I'm not sure" or "I don't know." We'll get there one day, and until then I'm happy to use it for the utility and convenience it provides while doing my part to make it better and more useful.
> Just because something didn't work out doesn't mean it was a waste
Here i think it's more about opportunity cost.
> I can't figure out what people mean when they say "AGI" any more, we appear to be past that
What i ask of an AGI is to not hallucinate idiotic stuff. I don't care about being bullshitted too much if the bullshit is logic, but when i ask "fix mypy errors using pydantic" and instead of declaring a type for a variable it invent weird algorithms that make no sense and don't work (and the fix would have taken 5 minutes for any average dev).I mean, Claude 4.5 and Codex have replaced my sed/search and replaces, write my sanity tests, write my commit comment, write my migration scripts (and most of my scripts), and make refactor so easy i now do one refactor every month or so, but if it is AGI, i _really_ wonder what people mean by intelligence.
> Also, if anyone wants to know what a real effort to waste a trillion dollars can buy
100% agree. Pleas Altman, Ilya and other, i will hapilly let you use whatever money you want if that money is taken from war profiteers and warmongers.
> Just because something didn't work out doesn't mean it was a waste
One thing to keep in mind, is that most of these people who go around spreading unfounded criticism of LLMs, "Gen-AI" and just generally AI aren't usually very deep into understanding computer science, and even less science itself. In their mind, if someone does an experiment, and it doesn't pan out, they'll assume that means "science itself failed", because they literally don't know how research and science work in practice.
Maybe true in general, but Gary Marcus is an experienced researcher and entrepreneur who’s been writing about AI for literally decades.
I’m quite critical, but I think we have to grant that he has plenty of credentials and understands the technical nature of what he’s critiquing quite well!
Yeah, my comment was mostly about the ecosystem at large, rather than a specific dig to this particular author, I mostly agree with your comment.
Eh, tearing down a straw man is not an impressive argument from you either.
As a counter-point, LLMs still do embarrassing amounts of hallucinations, some of which are quite hilarious. When that is gone and it starts doing web searches -- or it has any mechanisms that mimic actual research when it does not know something -- then the agents will be much closer to whatever most people imagine AGI to be.
Have LLMs learned to say "I don't know" yet?
> Have LLMs learned to say "I don't know" yet?
Can they, fundamentally, do that? That is, given the current technology.
Architecturally, they don't have a concept of "not knowing." They can say "I don't know," but it simply means that it was the most likely answer based on the training data.
A perfect example: an LLM citing chess rules and still making an illegal move: https://garymarcus.substack.com/p/generative-ais-crippling-a...
Heck, it can even say the move would have been illegal. And it would still make it.
> When that is gone and it starts doing web searches -- or it has any mechanisms that mimic actual research when it does not know something
ChatGPT and Gemini (and maybe others) can already perform and cite web searches, and it vastly improves their performance. ChatGPT is particularly impressive at multi-step web research. I have also witnessed them saying "I can't find the information you want" instead of hallucinating.
It's not perfect yet, but it's definitely climbing human percentiles in terms of reliability.
I think a lot of LLM detractors are still thinking of 2023-era ChatGPT. If everyone tried the most recent pro-level models with all the bells and whistles then I think there would be a lot less disagreement.
Well please don't include me in some group of Luddites or something.
I use the mainstream LLMs and I've noted them improving. They have ways to go still.
I was objecting to my parent poster's implication that we have AGI. However muddy that definition is, I don't feel like we do have that.
I believe in a very practical definition of AGI. AGI is a system capable of RSI. Why? Because it mimics humans. We have some behaviours that are given to us from birth, but the real power of humans is our ability to learn and improve ourselves and the environment around us.
A system capable of self improvement will be sufficient for AGI imo.
Ah - recursive self improvement. I was thinking repetitive strain injury was odd. But that's probably quite a good test although LLMs may be able to improve a bit but still not be very good. An interesting point for me is if all humans went away could the AI/robots keep on without us which would require them to be able to maintain and build power plants, chip fabs and the like. A way to go on that one.
Self improvement doesn’t mean self improvement in any possible direction without any tradeoffs. Genetic algorithms can do everything an LLM can given enough computational resources and training, but being wildly inefficient humanity can’t actually use them to make a chatbot on any even vaguely relevant timeline.
No suprise to see self-congratulations and more "I'm the only person who ever questioned genAI" nonsense as the key parts of this article. What a bore Marcus is.
There was a lot of talk about reaching "peak AI" in early summer of this year.
I guess there is some truth to it. The last big improvement to LLMs was reasoning. It gave the existing models additional capabilities (after some re-training).
We've reached the plateau of tiny incremental updates. Like with smartphones. I sometime still use an iPhone 6s. There is no fundamental difference compared to the most current iPhone generation 10 years later. The 6s is still able to perform most of the tasks you need a smartphone to do. The new ones do it much faster, and everything works better, but the changes are not disrupting at all.
"reasoning."
It's real reasoning. But it's not comparable to a human level.
That's like saying a trillion dollars was potentially wasted sending men to the moon. You have to close your eyes to so much obvious progress and dissect your idea beyond recognition to start believing this thesis.
The core argument here, as far as I can discern it, seems to be: A trillion dollars has been spent scaling LLMs in an attempt to create AGI. Since scaling alone looks like it won't produce AGI, that money has been wasted.
This is a frankly bizarre argument. Firstly, it presupposes that _only_ way AI becomes useful is if turns into AGI. But that isn't true: Existing LLMs can do a variety of economically valuable tasks, such as coding, even when not being AGI. Perhaps the economic worth of non-AGI will never equal what it costs to build an operate it, but it seems way too early to make that judgement and declare any non-AGI AI as worthless.
Secondly, even if scaling alone won't reach AGI, that doesn't mean that you can reach AGI _without_ scaling. Even when new and better architectures are developed, it still seems likely that, between two models with an equivalent architecture, the one with more data and compute research will be more powerful. And waiting for better architectures before you try to scale means you will never start. 50 years from now, researchers will have much better architectures. Does that mean we should wait 50 years before trying to scale them? How about 100 years? At what point do you say, we're never going to discover anything better, so now we can try scaling?
I really struggle to come up with a reason that transformers won't continue to deliver on additional capabilities that get fit into the training set.
companies are already wasting majority fractions of their engineering labor spend on coordination costs and fake work, through that lens i have trouble making an argument that any of this matters. Which is why they are able to do it. I’m reminded of an old essay arguing that the reason Google spends so lavishly is because if they only spent what they needed, they would appear so extraordinarily profitable that the government would intervene.
Don't research computations also require substantial hardware?
I think the article makes decent points but I don't agree with the general conclusion here, which is that all of this investment is wasted unless it "reaches AGI." Maybe it isn't necessary for every single dollar we spend on AI/LLM products and services to go exclusively toward the goal of "reaching AGI?" Perhaps it's alright if these dollars instead go to building out useful services and applications based on the LLM technologies we already have.
The author, for whatever reason, views it as a foregone conclusion that every dollar spent in this way is a waste of time and resources, but I wouldn't view any of that as wasted investment at all. It isn't any different from any other trend - by this logic, we may as well view the cloud/SaaS craze of the last decade as a waste of time. After all, the last decade was also fueled by lots of unprofitable companies, speculative investment and so on, and failed to reach any pie-in-the-sky Renaissance-level civilization-altering outcome. Was it all a waste of time?
It's ultimately just another thing industry is doing as demand keeps evolving. There is demand for building the current AI stack out, and demand for improving it. None of it seems wasted.
That’s not what he’s saying, the investors are the ones who have put trillions of dollars into this technology on the premise that it will achieve AGI. People like Sam Altman and Marc Andreesen have been going into podcasts saying AGI is imminent and they’re going to automate every job.
The author did not say every dollar was wasted, he said that LLMs will never meet the current investment returns.
It’s very frustrating to see comments like this attacking strawmans and setting up Motte and Bailey arguments every time there’s AI criticism. “Oh but LLMs are still useful” and “Even if LLMs can’t achieve AGI we’ll figure out something that will eventually.” Yes but that isn’t what Sam and Andreesen and all these VCs have been saying, and now the entire US economy is a big gamble on a technology that doesn’t deliver what they said it would and because the admin is so cozy with VCs we’re probably all going to suffer for the mistakes of a handful of investors who got blinded by dollar signs in their eyes.
Those billions of dollars spent on data centres could instead have hypothetically been spent on researchers with relatively modest computing resources. A few hundred billion dollars will buy you a lot of researchers.
The author quite literally says that the last few years were a "detour" that has wasted a trillion dollars. He explicitly lists building new LLMs, building larger LLMs and scaling LLMs as the problem and source of the waste. So I don't think I am strawmanning his position at all.
It is one thing to say that OpenAI has overpromised on revenues in the short term and another to say that the entire experiment was a waste of time because it hasn't led to AGI, which is quite literally the stance that Marcus has taken in this article.
> The author, for whatever reason, views it as a foregone conclusion that every dollar spent in this way is a waste of time and resources
This is a strawman, the author at no point says that “every dollar is a waste.”
He quite literally says that the dollars spent on scaling LLMs in the past few years are a waste.
Yes and you’re portraying it as “every dollar was wasted” and then arguing against that exaggerated claim which was never made.
It's the nature of bubbles to overpromise and underdeliver.
So - do Altman and Andreesen really believe that, or is it just a marketing and investment pitch?
A reminder that OpenAI has its own explicit definition of AGI: "Highly autonomous systems that outperform humans at most economically valuable work."
The MS/OAI agreement quantifies that as "profits of $100bn/yr."
This seems rather stupid to me. If you can get multiple AGIs all generating profits of $100bn a year - roughly half an Apple, or two thirds of a Meta - you no longer have anything recognisable as a pre-AI economy, because most of the white collar population is out of work and no longer earning anything. And most of the blue collar population that depends on white collar earnings is in the same boat.
So you have to ask "Profits from doing what, and selling to whom?"
The Altman pitch seems to be "Promise it first, collect investor cash, and worry about the rest later."
You are making the same strawman attack you are criticising.
The dollars invested are not justified considering TODAYs revenues.
Just like 2 years ago people said NVIDIA stock prices was not justified and a massive bubble considering the revenue from those days. But NVIDIA revenues 10xed, and now the stock price from 2 years ago looks seriously underpriced and a bargain.
You are assuming LLM revenues will remain flat or increase moderately and not explode.
You seem like someone who might be interested in my nuclear fusion startup. Right now all we have is a bucket of water but in five years that bucket is going to power the state of California.
It's not about "every dollar spent" being a waste of time, it's about acknowledging the reality of opportunity cost. Of course, no one in any movement is likely to listen to their detractors, but in this case the pioneers seem to agree.
https://www.youtube.com/watch?v=DtePicx_kFY https://www.bbc.com/news/articles/cy7e7mj0jmro
I think there is broad agreement that new models and architectures are needed, but I don't see it as a waste to also scale the stack that we currently have. That's what Silicon Valley has been doing for the past 50 years - scaling things out while inventing the next set of things - and I don't see this as any different. Maybe current architectures will go the way of the floppy disk, but it wasn't a waste to scale up production of floppy disk drives while they were relevant. And ChatGPT was still released only 3 years ago!
And notably, Marcus has been banging this drum for years. Even this article points back to articles he wrote years ago suggesting deep learning was hitting the wall... With GPT 3....
It's sour grapes because the methods he prefers have not gotten the same attention (hah...) or funding.
He's continuing to push the ludicrous Apple "reasoning paper" that he described as a "knockout blow for LLMs" even though it was nothing of the sort.
With each of his articles, I usually lose more respect for him.
It seems that innovators, researchers, and founders all work at a fast pace, but adoption of new technology, especially LLMs, ends up being done by companies as is convenient. When an open position can go unfilled or a group can scale up without hiring then companies might move forward with a commitment to LLMs.
Even with strong adoption it may take many years for LLMs available now to reach their potential utility in the economy. This should moderate the outlook for future changes, but instead we have a situation where the speculative MIT study that predicted "AI" could perform 12% of the work in the economy is widely considered to not only be accurate, but inevitable in the short term. How much time is needed dramatically changes calculations of potential and what might be considered waste.
Also worth keeping in mind that the Y2K tech bust left behind excess network capacity that ended up being useful later, but the LLM boom can be expected to leave behind poorly considered data centers full of burned out chips which is a very different legacy.
Ali Kadri’s The Accumulation of Waste: A Political Economy of Systemic Destruction comes to mind
https://lpeproject.org/events/the-accumulation-of-waste-a-po...
LLMs write all my code now and I just have to review it. Not only has my output 3x'ed at least, I also have zero hesitations now tackling large refactors, or tracking down strange bugs. For example, I recently received a report there was some minor unicode related data corruption in some of our doc in our DBs. It was cosmetic, and low priority, also not a simple task to track down traditionally. But now I just put [llm agent on it, to avoid people accusing me of promoting] on it. It found 3 instances of the corruption across hundreds of documents and fixed them.
I am sure some of you are thinking "that is all slop code". It definitely can be if you don't do your due diligence in review. We have definitely seen a bifurcation of devs who do that, and those who don't, where I am currently working.
But by far the biggest gain is my mental battery is far less drained at the end of the day. No task feels soul crushing anymore.
Personally, coding agents are the greatest invention of my lifetime outside the emergence of the internet.
Well those chips and power plants might still be useful for what comes after.
If we find AGI needs a different chip architecture, yeah, LLMs would have been quite a waste.
Did someone say that LLM was the final solution while I wasn’t listening? Am I fantasizing the huge outcry about the terrible danger of AGI? Are people not finding ways to use the current levels of LLM all over the place?
The idea that the trillions are a waste is not exactly fresh. The economic model is still not clear. Alarmists have been shrill and omnipresent. Bankruptcy might be the future of everyone.
But, will we look up one day and say, “Ah never mind” about GPT, Claude, et al? Fat chance. Will no one find a use for a ton of extra compute? I’m pretty sure.
I don’t much dispute any of the facts I skimmed off the article but the conclusion is dumb.
Ironically that MIT study that made the rounds a few months ago ("Study finds 90% of AI pilots fail" you remember), also found that virtually every single worker at every company they studied was using LLMs regularly.
The real takeaway of the study was that workers were using their personal LLM accounts to do work rather than using the AI implementation mess their companies had shat out.
Personally I think we'll find something better than the LLM algorithm fairly soon, but it will still be using the same GPU type servers.
If bankruptcy does happen to be the future for everyone, then yes, I think there is going to be a lot of "ah never mind"s going around.
If all this went away tomorrow, what would we do with all the compute? Its not exactly general purpose infrastructure thats being built.
My hypothesis is that general computing frameworks are the next big thing. The powerful GPUs have been mostly black boxes for way too long. A lot of clever people will not want to just throw them away or sell them second-hand and will try to find better ways to utilize them.
I might very well be super wrong. F.ex. NVIDIA is guarding their secrets very well and we have no reason to believe they'll suddenly drop the ball. But it does make me think; IMO a truly general GPU (and open + free) compute has been our area's blind spot for way too long.
Some of the participants may go bust but I very much doubt the highly profitable ones like Google, Apple, Nvidia and Microsoft will. There'll be enough demand for existing LLMs to keep the servers busy. Just writing code which works currently is probably enough to justify a fair chunk of the capacity.
Folding@Home?
Could always mine crypto.
I’m glad the 0.01% have something to burn their money on.
To further your point - I mean honestly if this all ends up being an actual bubble that doesn’t manifest a financial return for the liquidity injectors but instead a massive loss (for the .01% who are in large part putting the cash in), did humanity actually lose?
If it pops it might end up being looked at in the lens of history as one of the largest backdoor/proxy wealth redistributions ever. The capex being spent is in large part going to fund the labor of the unwashed masses, and society is getting the individual productivity and efficiency benefits from the end result models.
I’m particularly thankful for the plethora of open source models I have access to thanks to all this.
I, individually, have realized indisputable substantial benefits from having these tools at my disposal every day. If the whole thing pops, these tools are safely in my possession and I’m better because I have them. Thanks .01%!!
(the reality is I don’t think it will pop in the classic sense, and these days it seems the .01 can never lose. either way, the $1tn can’t be labeled as a waste).
It would be nice if they could burn it on something that didn't require them to buy up the world's supply of DDR5 RAM, and triple prices for everyone else.
https://pcpartpicker.com/trends/price/memory/
that might be literally the least of my concern regarding gen AI in today's world
Every technological change has been accompanied by an investment boom that resulted in some degree of wasted investment: cars, electricity, mass production of bicycles, it goes on and on.
One point about this is that humans appear unable to understand that this is an efficient outcome because investment booms are a product of uncertainty around the nature of the technological change. You are building something is literally completely new, no-one had any idea what cars consumers would buy so lots of companies started to try and work out that out and that consolidated into competition on cost/scale once that became clear. There is no way to go to the end of that process, there are many people outside the sphere of business who are heavily incentivized to say that we (meaning bureaucrats and regulators) actually know what kind of cars consumers wanted and that all the investment was just a waste.
Another point is that technological change is very politically disruptive. This was a point that wasn't well appreciated...but is hopefully clear with social media. There are a large number of similar situations in history though: printing press, newspapers, etc. Technological change is extremely dangerous if you are a politician or regulator because it results in your power decreasing and, potentially, your job being lost. Again, the incentives are huge.
The other bizarre irony of this is that people will look at an investment boom with no technological change, that was a response to government intervention in financial markets and a malfunctioning supply-side economy...and the response was: all forms of technical innovation are destabilizing, investment booms are very dangerous, etc. When what they mean is corporations with good political connections might lose money.
This is also linked inherently to the view around inflation. The 1870s are regarded as one of the most economically catastrophic periods in economic history by modern interpretations of politics. Let me repeat this in another way: productivity growth was increasing by 8-10%/year, you saw mind-boggling gains from automation (one example is cigarettes, iirc it took one skilled person 10-20 minutes to create a cigarette, a machine was able to produce hundreds in a minute), and conventional macroeconomics views this as bad because...if you can believe it...they argue that price declines lead to declines in investment. Now compare to today: prices continue to rise, investment is (largely) non-existent, shortages in every sector. Would you build a factory in 1870 knowing you could cut prices for output by 95% and produce more? The way we view investment is inextricably linked in economic policy to this point of view, and is why the central banks have spent trillions buying bonds with, in most cases, zero impact on real investment (depending on what you mean, as I say above, private equity and other politically connected incumbents have made out like bandits...through the cycle, the welfare gain from this is likely negative).
You see the result of this all over the Western world: shortages of everything, prices sky-high, and when technological change happens the hysteria around investment being wasteful and disruptive. It would be funny if we didn't already see the issues with this path all around us.
It is not wasted, we need more of this, this ex-post, academic-style reasoning of everything in hindsight gets us nowhere. There is no collateral damage, even in the completely fake Fed-engineered housing bubble, the apparently catastrophic cost was: more houses, and some wealthy people lost their net worth (before some central bankers found out their decisions in 03-04 caused wealthy people to lose money, and quickly set about recapitalising their brokerage accounts with taxpayers money).
Nothing new here, just nepo as old as time.
Perhaps the scale is unprecedented, or it's always been like this it's just much less concealed these days.
Absolute retards can waste trillions of dollars on stupid ideas, because they're in the in group. Next door someone who's worked their whole life gets evicted because their mortgage is now way more of what they make in salary.
Sucks to be in the out group!
I always love a Marcus hot take, but this one is more infuriating than usual. He’s taking all these prominent engineers saying “we need new techniques to build upon the massive, unexpected success we’ve had”, twisting it into “LLMs were never a success and sucked all along”, and listing them alongside people that no one should be taking seriously — namely, Emily Bender and Ed Zitron.
Of course, he includes enough weasel phrases that you could never nail him down on any particular negative sentiment; LLMs aren’t bad, they just need to be “complemented”. But even if we didn’t have context, the whole thesis of the piece runs completely counter to this — you don’t “waste” a trillion dollars on something that just needs to be complemented!
FWIW, I totally agree with his more mundane philosophical points about the need to finally unify the work of the Scruffies and the Neats. The problem is that he frames it like some rare insight that he and his fellow rebels found, rather than something that was being articulated in depth by one of the fields main leaders 35 years ago[1]. Every one of the tens of thousands of people currently working on “agential” AI knows it too, even if they don’t have the academic background to articulate it.
I look forward to the day when Mr. Marcus can feel like he’s sufficiently won, and thus get back to collaborating with the rest of us… This level of vitriolic, sustained cynicism is just antithetical to the scientific method at this point. It is a social practice, after all!
[1] https://www.mit.edu/~dxh/marvin/web.media.mit.edu/~minsky/pa...
[dead]
[dead]
“He is not forecasting a bright future for LLMs”.
Yeah, no shit. I’ve been saying this since the day GPT 3 became hyped. I don’t think many with a CS background are buying the “snake oil” of AGI through stochastic parrots.
At some point, even people who hype LLMs will spin their narrative to not look out of touch with reality. Or not more out of touch than is acceptable lol.
A nice study on the parrot snake oil
https://ai.vixra.org/pdf/2506.0065v1.pdf
When it comes to machine learning, research has consistently shown, that pretty much the only thing that matters is scaling.
Ilya should just enjoy his billions raised with no strings.
If you think scaling is all that matters, you need to learn more about ML.
Read about the the No Free Lunch Theorem. Basically, the reason we need to "scale" so hard is because we're building models that we want to be good at everything. We could build models that are as good at LLMs at a narrow fraction of tasks we ask of them to do, at probably 1/10th the parameters.
Are reranker models an example of this? Do they still underperform compared to LLMs?
> When it comes to machine learning, research has consistently shown, that pretty much the only thing that matters is scaling.
Yes, indeed, that is why all we have done since the 90s is scale up the 'expert systems' we invented ...
That's such an a-historic take it's crazy.
* 1966: failure of machine translation
* 1969: criticism of perceptrons (early, single-layer artificial neural networks)
* 1971–75: DARPA's frustration with the Speech Understanding Research program at Carnegie Mellon University
* 1973: large decrease in AI research in the United Kingdom in response to the Lighthill report
* 1973–74: DARPA's cutbacks to academic AI research in general
* 1987: collapse of the LISP machine market
* 1988: cancellation of new spending on AI by the Strategic Computing Initiative
* 1990s: many expert systems were abandoned
* 1990s: end of the Fifth Generation computer project's original goals
Time and time again, we have seen that each academic research begets a degree of progress, improved by the application of hardware and money, but ultimately only a step towards AGI, which ends with a realisation that there's a missing congitive ability that can't be overcome by absurd compute.
LLMs are not the final step.
Well, expert systems aren’t machine learning, they’re symbolic. You mention perceptrons, but that timeline is proof for the power of scaling, not against — they didn’t start to really work until we built giant computers in the ~90s, and have been revolutionizing the field ever since.
Indeed. This is the "bitter lesson".
https://en.wikipedia.org/wiki/Bitter_lesson
Didn’t OpenAI themselves publish a papers years ago that scaling parameters has diminishing returns?