The AI boom has turned the standard profit margin model on its head, according to Apollo Chief Economist Torsten Slok—and it’s making the industry’s growth unsustainable.
The issue is that AI companies have been funded on the promise that with enough advancements and upgrades this tech will achieve AGI. Which is an absurd statement to anyone in the field actually developing these things. That is the equivalent of promising that this 260mph Bugatti, with a few years of upgrades and advancements, will become a Harrier Jump Jet!
A better analogy is promising that a toddler will, with enough knowledge and training, eventually become a heart surgeon.
15 years ago the assumption with AGI was that we needed some paradigm shift in technology to create it, but after transformer technology was invented (the T in GPT), and it was trained on a lot of information, we discovered an emergent property that it could take natural language queries and answer them with its knowledge base-- which was unexpected and unintended.
While science can always end up going down the wrong path, the current mainstream stance is that we were wrong about needing new technology for AGI; it seems that AGI may be a function of information and training, on hardware we already have. Hence all the data centers being built.
Anyone who says with certainty that it will result in AGI is just as wrong as someone who says with certainty that it won’t.
That is where the evidence points. Now, I don’t want to oversell it: “where the evidence points” is wildly different than “exactly how it works”.
We have an emergent property that we don’t understand, but we can reliably increase the functionally and complexity of that emergent property as a function of training data and available compute. Does that mean that there isn’t some threshold where that stops working? No, there certainly could be a point where throwing more information and compute has no effect. We just don’t know. However, so far, there is no evidence such a barrier exists, and everyone is racing to find out.
I think that is where the hopes and hype point. The evidence, that which is gathered through controlled studies, points to an upper limit to this technology that does have emergent properties, but not ones that amount to cognition. That evidence also points to other hurdles, such as cognitive damage to the user and context windows nowhere near that of even a simpleminded creature, let alone a sapient one like a human.
The core problem is that these models are fixed; they are the same on day 1000 as they were on day 1. All their “learning”, as it were, happens in training before it is released. Everything it appears to learn after that date is contained in the rolling context window. Since they already have literally all the RAM they can get their hands on and are still at least an order of magnitude away from where they need to be on that, this technology either can’t do it or, at best, is so inefficient an approach that it can’t be done with all the planet’s resources.
Sometimes, especially in abstract constructions like software, you can start down the wrong path early and have to start over, because there is no path from where you are to where you need to get. In this case, they may have done that to the extreme, blinded by the lucrative prospects.
As far as I am aware, AGI does not imply sapience, consciousness, or whatever. You seem to believe it does. A high level Google search seems to suggest that I’m correct. Is there some reason you are lumping sapience with AGI?
I’m not sure if the rest of your comment presupposes this, so I’ll wait to comment on where I think you’re mistaken.
I dont believe it requires sapience; that is what the marketers are saying. The AI boom (and many historic boom cycles) is predicated on marketing and sentiment, not facts and data. That is why they always pop; the facts dont back up the hype.
The fact that your search results turn up results that align with the marketing is just the marketing working.
Look–neither of us are data scientists, but we do have eyes. If this is working, where’s the company with runaway success creating unimagined leaps in productivity and technology? We’re pouring a whole planet’s with of resources in and nothing much is coming out. If we spent this much on world hunger, everyone would be obese by now.
I dont believe it requires sapience; that is what the marketers are saying. The AI boom (and many historic boom cycles) is predicated on marketing and sentiment, not facts and data. That is why they always pop; the facts dont back up the hype.
I think there is some confusion. I do not believe AGI implies sapience, nor does Google, and now it seems that you don’t either. So who is discussing sapience?
If this is working, where’s the company with runaway success creating unimagined leaps in productivity and technology?
Hypothetically speaking, what do you think this would look like?
The marketers are not using the word “sapience”, as that word would require defining for the layman. They are claiming AI will usher in an era of human prosperity by taking over our work and big decisions to do them more efficiently–tasks that sound to almost everyone to require advanced decision-making and thought, the hallmarks of sapience and cognition. This is largely besides the point, though.
I think that if this were successful, you would see at least some early adopting companies coming up aces. Some company would have double the productivity with half the workforce, and use that to absolutely devour their market segment by undercutting everyone else on price because their operating costs are so much lower than the competition. I’d see a software company adopt breakneck software release cycles with substantial, material stability and performance improvements with each. I’d see some company anywhere suddenly begin to outperform their former selves and be able to convincingly point to AI as where their improvements came from. The only one like that I can think of is NVidia, who are rich strictly because they’re selling shovels in the gold rush.
What it would look like is layoffs from a company without a change in output/productivity. Notably, we’re not at AGI yet, so I wouldn’t expect sci-fi levels of change.
Vishal Sikka, advisory board member of BMW. Recommended to Stanford by Marvin Minsky, one of his professors were John McCarty. And I must stand corrected, he has a PHD of computer sciences, not Math as I remembered it as.
Vishal has an AI based company himself, so there might be some personal reasons for why he’d advocate for using what AIs capable of rather than chasing an impossible (from his perspective) to hit milestone
That paper doesn’t seem to rule out AGI, only an single LLM model that can answer every arbitrarily difficult question on demand.
AGI does not necessarily mean one model acting alone, or being able to answer any question on demand. Humans are the same way: we often need time or collaboration to arrive at conclusions, but that doesn’t mean we don’t have “general intelligence”.
You’re not going to context hack complexity, every time you summarise something and pass it onto the next agent. You’re losing complexity.
We’re going to get some massive models with incredible context, but AGI requires models to never hallucinate. Even when the complexity requires more context than it has available, which is not feasible
We’ve already had a month now of Psychiatrists writing articles backpedaling over all the times they’d insisted AI couldn’t genuinely be thinking as we do and moving the goal posts to something else. Maybe there are researchers in the frontier labs who feel like they have a grip on what’s happening, but the ‘experts’ on forums have no idea.
A better analogy is promising that a toddler will, with enough knowledge and training, eventually become a heart surgeon.
15 years ago the assumption with AGI was that we needed some paradigm shift in technology to create it, but after transformer technology was invented (the T in GPT), and it was trained on a lot of information, we discovered an emergent property that it could take natural language queries and answer them with its knowledge base-- which was unexpected and unintended.
While science can always end up going down the wrong path, the current mainstream stance is that we were wrong about needing new technology for AGI; it seems that AGI may be a function of information and training, on hardware we already have. Hence all the data centers being built.
Anyone who says with certainty that it will result in AGI is just as wrong as someone who says with certainty that it won’t.
That analogy makes the assumption that LLMs are dormant baby AGIs. There is no evidence that AGI is just a bigger and better curated LLM.
That is where the evidence points. Now, I don’t want to oversell it: “where the evidence points” is wildly different than “exactly how it works”.
We have an emergent property that we don’t understand, but we can reliably increase the functionally and complexity of that emergent property as a function of training data and available compute. Does that mean that there isn’t some threshold where that stops working? No, there certainly could be a point where throwing more information and compute has no effect. We just don’t know. However, so far, there is no evidence such a barrier exists, and everyone is racing to find out.
I think that is where the hopes and hype point. The evidence, that which is gathered through controlled studies, points to an upper limit to this technology that does have emergent properties, but not ones that amount to cognition. That evidence also points to other hurdles, such as cognitive damage to the user and context windows nowhere near that of even a simpleminded creature, let alone a sapient one like a human.
The core problem is that these models are fixed; they are the same on day 1000 as they were on day 1. All their “learning”, as it were, happens in training before it is released. Everything it appears to learn after that date is contained in the rolling context window. Since they already have literally all the RAM they can get their hands on and are still at least an order of magnitude away from where they need to be on that, this technology either can’t do it or, at best, is so inefficient an approach that it can’t be done with all the planet’s resources.
Sometimes, especially in abstract constructions like software, you can start down the wrong path early and have to start over, because there is no path from where you are to where you need to get. In this case, they may have done that to the extreme, blinded by the lucrative prospects.
As far as I am aware, AGI does not imply sapience, consciousness, or whatever. You seem to believe it does. A high level Google search seems to suggest that I’m correct. Is there some reason you are lumping sapience with AGI?
I’m not sure if the rest of your comment presupposes this, so I’ll wait to comment on where I think you’re mistaken.
I dont believe it requires sapience; that is what the marketers are saying. The AI boom (and many historic boom cycles) is predicated on marketing and sentiment, not facts and data. That is why they always pop; the facts dont back up the hype.
The fact that your search results turn up results that align with the marketing is just the marketing working.
Look–neither of us are data scientists, but we do have eyes. If this is working, where’s the company with runaway success creating unimagined leaps in productivity and technology? We’re pouring a whole planet’s with of resources in and nothing much is coming out. If we spent this much on world hunger, everyone would be obese by now.
I think there is some confusion. I do not believe AGI implies sapience, nor does Google, and now it seems that you don’t either. So who is discussing sapience?
Hypothetically speaking, what do you think this would look like?
The marketers are not using the word “sapience”, as that word would require defining for the layman. They are claiming AI will usher in an era of human prosperity by taking over our work and big decisions to do them more efficiently–tasks that sound to almost everyone to require advanced decision-making and thought, the hallmarks of sapience and cognition. This is largely besides the point, though.
I think that if this were successful, you would see at least some early adopting companies coming up aces. Some company would have double the productivity with half the workforce, and use that to absolutely devour their market segment by undercutting everyone else on price because their operating costs are so much lower than the competition. I’d see a software company adopt breakneck software release cycles with substantial, material stability and performance improvements with each. I’d see some company anywhere suddenly begin to outperform their former selves and be able to convincingly point to AI as where their improvements came from. The only one like that I can think of is NVidia, who are rich strictly because they’re selling shovels in the gold rush.
What it would look like is layoffs from a company without a change in output/productivity. Notably, we’re not at AGI yet, so I wouldn’t expect sci-fi levels of change.
We have math PHDs with proofs current LLMs can’t become AGI, because they’ll always have a context issue
I would love it if you’d point me at these mathematical proofs.
This is the primary paper I reference.
https://arxiv.org/pdf/2507.07505
Vishal Sikka, advisory board member of BMW. Recommended to Stanford by Marvin Minsky, one of his professors were John McCarty. And I must stand corrected, he has a PHD of computer sciences, not Math as I remembered it as.
Varin Sikka is his son, co author of the paper and based on Stanford’s site an undergraduate. https://profiles.stanford.edu/363374
Vishal has an AI based company himself, so there might be some personal reasons for why he’d advocate for using what AIs capable of rather than chasing an impossible (from his perspective) to hit milestone
That paper doesn’t seem to rule out AGI, only an single LLM model that can answer every arbitrarily difficult question on demand.
AGI does not necessarily mean one model acting alone, or being able to answer any question on demand. Humans are the same way: we often need time or collaboration to arrive at conclusions, but that doesn’t mean we don’t have “general intelligence”.
You’re not going to context hack complexity, every time you summarise something and pass it onto the next agent. You’re losing complexity.
We’re going to get some massive models with incredible context, but AGI requires models to never hallucinate. Even when the complexity requires more context than it has available, which is not feasible
We’ve already had a month now of Psychiatrists writing articles backpedaling over all the times they’d insisted AI couldn’t genuinely be thinking as we do and moving the goal posts to something else. Maybe there are researchers in the frontier labs who feel like they have a grip on what’s happening, but the ‘experts’ on forums have no idea.