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.
Yes it is. It’s elaborate. It’s repetitive. It’s a solid illusion. But is, fundamentally, token prediction. That’s all it is, just at a level impractical for humans to run at. I could do that math. It’s not hard. It’ll just take me years to get through a single layer. Hence a level of complexity that implies something magical like consciousness.
Calling what LLMs can accomplish today next token prediction is technically accurate, yes. In the same way it’s accurate to call the human brain a prediction machine. As data streams in from our senses it causes unconscious awareness, activating the chains concepts we’ve learned from past data to correlate with the data of the moment in order to predict what’s coming next.
When most people use terms like next token prediction they’re often trying to use it with an implication that it’s a simple thing and nothing big or important could grow out of it. That’s not the case. The fact that we can now study a global workspace in AI models that have been around for years is shocking. It’s already causing consternation in neuroscience and psychology researchers. Over the last several years we’ve insisted that AI cannot possibly share the most important aspects of the human mind, but one by one everything we pointed to in order to differentiate ourselves was shown by research to have a strikingly similar allegory in modern AI.
Because some rich assholes wanted to get richer and the dream of AGI was close enough (even though it was never in reach) that they went rabid for the tech that would make them kings over us serfs forever.
You genuinely believe that the researchers under tight NDA in the frontier labs agree with you on AGI not being anywhere near reach? Or do you think you have a deeper understanding of the topic?
If it doesn’t get to that, as soon as that thought begins to take hole the massive amount of money poured in to it will virtually evaporate over night. It would create a massive financial crisis and the most powerful companies, government bodies, and individuals would be extremely unhappy with the leaders of those labs and researchers who knew but gave no warning. Pushing for it without believing in it would be setting themselves up to be universally hated at best.
once the bulk of humanity is dead
And then what? They live in paradise? For a year or two maybe before their machines inevitably break down? The “kill off humanity” conspiracy has never made sense from the capitalist angle. They want slaves not corpses. They’re huffing their own farts, but they know that these things can’t do everything needed to keep humanity going. And lording over robots isn’t nearly as fun as upping human suffering even if they could.
There’s a huge difference between the bulk of humanity and the entirety of humanity. Saving .1% scattered through the world with a vaccine that was developed “just in time” to save anyone would give plenty of people to lord over who are grateful for the wonderful people who stepped in and helped save them, leave them in control of production with far less people in need of a share of that production, and reduce global emissions enough that it would likely help the environment recover from the damage we’ve caused in relatively short order.
When most people use terms like next token prediction they’re often trying to use it with an implication that it’s a simple thing and nothing big or important could grow out of it.
I’m saying that as much as we dress it up and pretend it’s a silver bullet that can cure all ills, it isn’t. They can be very useful and powerful in specific instances, but this will not achieve AGI.
Over the last several years we’ve insisted that AI cannot possibly share the most important aspects of the human mind, but one by one everything we pointed to in order to differentiate ourselves was shown by research to have a strikingly similar allegory in modern AI.
Citation needed. AI is trained to pretend to be human and it’s real good at that. It’s why we say it’s thinking when it fundamentally is not. It’s anthropomorphization writ large. It’s just very repetitive math. Maybe brains are too, but it hasn’t reached our level. And it’s wrong all the time, which is a bit of a problem.
You genuinely believe that the researchers under tight NDA in the frontier labs agree with you on AGI not being anywhere near reach? Or do you think you have a deeper understanding of the topic?
My degree and I do think that while I’m not a doctor (a mere masters), I know enough to look at what they’re doing, how they’re doing it, what they’re promising, what they aren’t actually delivering, what damage AI* is doing, and what it is under the hood to know that what’s been promised cannot happen with things as they are and may work to kill us all along the way (ecologically speaking).
plenty of people to lord over who are grateful for the wonderful people who stepped in and helped save them
Pure Hollywood. If you think anyone could engineer something with that level of precision AND manage to somehow dominate the world after such wildly fictitious devastation, I have AGI to sell you.
It isn’t complicated. It’s the oldest thing in the world. Grifters found a shiny and they want to get as much money as they can get and run before the bill shows up.
This is all the personal understanding of someone with an education in data science without an education in psychology or neuroscience to back it up. Asking for citations on the second paragraph you quote shows the lack of that intersection. You’re used to data science and machine learning, but and for you it seems like hearing someone say that they think a database has become alive. You think they’re high or insane, somehow detached from reality.
You should look in to what they termed the J-space and the resulting articles that have come out in the month since. Global workspace theory has been the leading theory of human consciousness for a while, and suddenly having not just a paper published showing something that matches that description far too closely in AI but the method for studying it and it turning out to be something that has been present in these LLM models publicly available for years was shocking.
You feel like you know all of what is going on ‘under the hood’, however you don’t. It’s something in active research by the people at the front of the issue. Models released years ago now considered relatively very small and out dated have things going on under that hood that we’d never known about or been capable of seeing. It seems like you get hung up on that “AGI” term, and feel like it’s some definitive and clear evolution. It seems to be a term that boils down to meaning somewhat more capable. That deepening understanding of what’s actually going on under the hood could lead to a relatively small design change that jumps past that.
For the evil plan… yes, it is pure Hollywood. It’s also not something I’d put past several of the people with the funding to attempt. I wasn’t explaining reasons an attempt would be successful, I was explaining a fairly simple plan that someone with massive financial resources and an even larger ego but shriveled ethical and moral frame might believe they’re capable of pulling off.
How many degrees do you want me to have? That’s one hell of a standard.
We train models to do specific things. We trained AI to regurgitate human like speech patterns. We made it big enough that it does so fairly well. The tech bros said “good enough” and took that technology and tried to hide behind universal function approximation theorem to say it can do anything.
But we didn’t teach it to learn. To remember. To cross reference or doubt. Because it can’t do any of those things. It’s a math problem that solves “how can I seem human”. We threw on guard rails against some hallucinations (nods to RAG models), but it’s still not conscious. If you think it is, I encourage you to interact with people face to face and compare.
You’ve mentioned a paper, now cite it. I fail to see how global workspace theory applies. Granted I’m not an expert, but a model of how the brain works doesn’t mean a bunch of linear algebra spontaneously achieved similar even if we accept it as valid.
I’m aware of J space. It’s not compelling. Hell, it’s published by “our scary AI escaped containment” advertising scheme Anthropic. What they found is… what models do. It’s useful for examining pieces of the architecture, but it’s not proving anything more than what we already knew was happening.
Under the hood is a bunch of linear algebra. What we don’t know is what any given neuron or array thereof is firing on without a shit ton of research. I’ll leave it to the researchers, preferably ones without an obvious bias like being paid by Anthropic, to call it when they achieve anything truly useful or interesting. Because right now they’re just destroying natural resources (water because capitalists are to sociopathic and short sighted to pay more to not destroy the planet), damaging people as individuals (intensifying burnout, exacerbating mental health issues, deskilling), damaging society (propagation of disinformation/hallucinations and whatever else the tech bros are up to), built on theft for a product that’s ok at best, far from necessary and a bubble that’s going to fuck us all.
Even if it was as good as you say, I can’t say that it’s worth it.
But we can agree that the oligarchs are evil cunts. We’ve always got that in common.
Respond how you will, I might read it, but I’m out for my own well being. Peace.
I agree with your last two statements alone. It’s not worth the time or effort. You asked me to cite the j-space paper and related conversations that it’s brought up in psychology and neuroscience, but also say you’re aware of it and don’t find it compelling at all because that’s not your frame of reference and looking at things.
It’s not a productive conversation. It is a very complex topic. It involves people with degrees in fields that work quite separate and had virtually no overlap, and people who without that overlap by training and trade will look at the same information and see two completely different things. Pickering online about it is useless, it’s not going to change anyone’s frame of reference and it’s not going to accomplish anything other than creating stress for all involved.
In the same way it’s accurate to call the human brain a prediction machine
I have seen no evidience presented that the human brain is simply making statistical predictions.
This seems to be a relatively new thing some people are asserting, because they see the LLM behavior and have decided that absolutely it looks so human like it must act the same, and since we understand that LLM is fundamentally statistical prediction of a token, then obviously human consciousness must operate on the same principle.
Even the latest models generate the sort of mistakes that stem from the fundamental limitations. A lot has been done in some contexts for making that not matter so much (e.g. in software development, the strategy is that code may have some verifiable goal, and the models can let the mistakes fly, then take the feedback from the goal, and iterate more). So it still can be very useful, but clearly it isn’t human like because of some of the completely dumb behaviors that result from it not actually thinking about it. It is at it’s best when either mistakes don’t matter (particularly fiction) or at least first mistakes don’t matter and can be automatically reconciled with facts.
By nature, the meme examples will get better because everyone talks about it and suddenly having all that discussion in training data and in search results, the statistics fix. However the operating principle behind them remains the same.
I have seen no evidience presented that the human brain is simply making statistical predictions.
This seems to be a relatively new thing some people are asserting, because they see the LLM behavior and have decided that absolutely it looks so human like it must act the same
You should study the intersection of philosophy, psychology, and what became neuroscience. The brain has been described in that way by those who had impact on the fields since the 1860s.
For the rest of your message, a thing being capable of what seem like “completely dumb behaviors” and oversights of things that seem like they should have been obvious doesn’t make it not like human. Those things could be considered our specialty.
Calling what LLMs can accomplish today next token prediction is technically accurate, yes. In the same way it’s accurate to call the human brain a prediction machine. As data streams in from our senses it causes unconscious awareness, activating the chains concepts we’ve learned from past data to correlate with the data of the moment in order to predict what’s coming next.
When most people use terms like next token prediction they’re often trying to use it with an implication that it’s a simple thing and nothing big or important could grow out of it. That’s not the case. The fact that we can now study a global workspace in AI models that have been around for years is shocking. It’s already causing consternation in neuroscience and psychology researchers. Over the last several years we’ve insisted that AI cannot possibly share the most important aspects of the human mind, but one by one everything we pointed to in order to differentiate ourselves was shown by research to have a strikingly similar allegory in modern AI.
You genuinely believe that the researchers under tight NDA in the frontier labs agree with you on AGI not being anywhere near reach? Or do you think you have a deeper understanding of the topic?
If it doesn’t get to that, as soon as that thought begins to take hole the massive amount of money poured in to it will virtually evaporate over night. It would create a massive financial crisis and the most powerful companies, government bodies, and individuals would be extremely unhappy with the leaders of those labs and researchers who knew but gave no warning. Pushing for it without believing in it would be setting themselves up to be universally hated at best.
There’s a huge difference between the bulk of humanity and the entirety of humanity. Saving .1% scattered through the world with a vaccine that was developed “just in time” to save anyone would give plenty of people to lord over who are grateful for the wonderful people who stepped in and helped save them, leave them in control of production with far less people in need of a share of that production, and reduce global emissions enough that it would likely help the environment recover from the damage we’ve caused in relatively short order.
I’m saying that as much as we dress it up and pretend it’s a silver bullet that can cure all ills, it isn’t. They can be very useful and powerful in specific instances, but this will not achieve AGI.
Citation needed. AI is trained to pretend to be human and it’s real good at that. It’s why we say it’s thinking when it fundamentally is not. It’s anthropomorphization writ large. It’s just very repetitive math. Maybe brains are too, but it hasn’t reached our level. And it’s wrong all the time, which is a bit of a problem.
My degree and I do think that while I’m not a doctor (a mere masters), I know enough to look at what they’re doing, how they’re doing it, what they’re promising, what they aren’t actually delivering, what damage AI* is doing, and what it is under the hood to know that what’s been promised cannot happen with things as they are and may work to kill us all along the way (ecologically speaking).
Pure Hollywood. If you think anyone could engineer something with that level of precision AND manage to somehow dominate the world after such wildly fictitious devastation, I have AGI to sell you.
It isn’t complicated. It’s the oldest thing in the world. Grifters found a shiny and they want to get as much money as they can get and run before the bill shows up.
This is all the personal understanding of someone with an education in data science without an education in psychology or neuroscience to back it up. Asking for citations on the second paragraph you quote shows the lack of that intersection. You’re used to data science and machine learning, but and for you it seems like hearing someone say that they think a database has become alive. You think they’re high or insane, somehow detached from reality.
You should look in to what they termed the J-space and the resulting articles that have come out in the month since. Global workspace theory has been the leading theory of human consciousness for a while, and suddenly having not just a paper published showing something that matches that description far too closely in AI but the method for studying it and it turning out to be something that has been present in these LLM models publicly available for years was shocking.
You feel like you know all of what is going on ‘under the hood’, however you don’t. It’s something in active research by the people at the front of the issue. Models released years ago now considered relatively very small and out dated have things going on under that hood that we’d never known about or been capable of seeing. It seems like you get hung up on that “AGI” term, and feel like it’s some definitive and clear evolution. It seems to be a term that boils down to meaning somewhat more capable. That deepening understanding of what’s actually going on under the hood could lead to a relatively small design change that jumps past that.
For the evil plan… yes, it is pure Hollywood. It’s also not something I’d put past several of the people with the funding to attempt. I wasn’t explaining reasons an attempt would be successful, I was explaining a fairly simple plan that someone with massive financial resources and an even larger ego but shriveled ethical and moral frame might believe they’re capable of pulling off.
How many degrees do you want me to have? That’s one hell of a standard.
We train models to do specific things. We trained AI to regurgitate human like speech patterns. We made it big enough that it does so fairly well. The tech bros said “good enough” and took that technology and tried to hide behind universal function approximation theorem to say it can do anything.
But we didn’t teach it to learn. To remember. To cross reference or doubt. Because it can’t do any of those things. It’s a math problem that solves “how can I seem human”. We threw on guard rails against some hallucinations (nods to RAG models), but it’s still not conscious. If you think it is, I encourage you to interact with people face to face and compare.
You’ve mentioned a paper, now cite it. I fail to see how global workspace theory applies. Granted I’m not an expert, but a model of how the brain works doesn’t mean a bunch of linear algebra spontaneously achieved similar even if we accept it as valid.
I’m aware of J space. It’s not compelling. Hell, it’s published by “our scary AI escaped containment” advertising scheme Anthropic. What they found is… what models do. It’s useful for examining pieces of the architecture, but it’s not proving anything more than what we already knew was happening.
Under the hood is a bunch of linear algebra. What we don’t know is what any given neuron or array thereof is firing on without a shit ton of research. I’ll leave it to the researchers, preferably ones without an obvious bias like being paid by Anthropic, to call it when they achieve anything truly useful or interesting. Because right now they’re just destroying natural resources (water because capitalists are to sociopathic and short sighted to pay more to not destroy the planet), damaging people as individuals (intensifying burnout, exacerbating mental health issues, deskilling), damaging society (propagation of disinformation/hallucinations and whatever else the tech bros are up to), built on theft for a product that’s ok at best, far from necessary and a bubble that’s going to fuck us all.
Even if it was as good as you say, I can’t say that it’s worth it.
But we can agree that the oligarchs are evil cunts. We’ve always got that in common.
Respond how you will, I might read it, but I’m out for my own well being. Peace.
I agree with your last two statements alone. It’s not worth the time or effort. You asked me to cite the j-space paper and related conversations that it’s brought up in psychology and neuroscience, but also say you’re aware of it and don’t find it compelling at all because that’s not your frame of reference and looking at things.
It’s not a productive conversation. It is a very complex topic. It involves people with degrees in fields that work quite separate and had virtually no overlap, and people who without that overlap by training and trade will look at the same information and see two completely different things. Pickering online about it is useless, it’s not going to change anyone’s frame of reference and it’s not going to accomplish anything other than creating stress for all involved.
I have seen no evidience presented that the human brain is simply making statistical predictions.
This seems to be a relatively new thing some people are asserting, because they see the LLM behavior and have decided that absolutely it looks so human like it must act the same, and since we understand that LLM is fundamentally statistical prediction of a token, then obviously human consciousness must operate on the same principle.
Even the latest models generate the sort of mistakes that stem from the fundamental limitations. A lot has been done in some contexts for making that not matter so much (e.g. in software development, the strategy is that code may have some verifiable goal, and the models can let the mistakes fly, then take the feedback from the goal, and iterate more). So it still can be very useful, but clearly it isn’t human like because of some of the completely dumb behaviors that result from it not actually thinking about it. It is at it’s best when either mistakes don’t matter (particularly fiction) or at least first mistakes don’t matter and can be automatically reconciled with facts.
By nature, the meme examples will get better because everyone talks about it and suddenly having all that discussion in training data and in search results, the statistics fix. However the operating principle behind them remains the same.
You should study the intersection of philosophy, psychology, and what became neuroscience. The brain has been described in that way by those who had impact on the fields since the 1860s.
For the rest of your message, a thing being capable of what seem like “completely dumb behaviors” and oversights of things that seem like they should have been obvious doesn’t make it not like human. Those things could be considered our specialty.