I have no idea how people can consider this to be a hype bubble especially after the o3 release. It smashed the ARC AGI benchmark on the performance front. It ranks as the 175th best competitive coder in the world on Codeforces’ leaderboard.
o3 proved that it is possible to have at least an expert AGI if not a Virtuoso AGI (according to Deep mind’s definition of AGI). Sure, it’s not economical yet. But it will get there very soon (just like how the earlier GPTs were a lot dumber and took a lot more energy than the newer, smaller parameter models).
Please remember - fight to seize the means of production. Do not fight the means of production themselves.
It’s a bubble because OpenAI spend $2.35 for every $1.00 they make. Yes, you’re mathing right, that is a net loss.
It’s a bubble because all of the big players in AI development agree that future models will cost exponentially more money to train, for incremental gains. That means there is no path forward that doesn’t intensely amplify the unprofitability of an already deeply unprofitable industry.
It’s a bubble because newer models with better capabilities only cost more and more to run.
It’s a bubble because as far as anyone knows there will never be a solution to the hallucination problem.
It’s a bubble because despite investments treating it as a trillion dollar industry, no one has yet figured out a trillion dollar problem that AI can solve.
You’re trying on a new top of the line VR headset and saying “Wow, this is incredible, how can anyone say this is a bubble?” Its not about how cool the tech is in isolation, it’s about its potential to effect widespread change. Facebook went in hard on VR, imagining a future where everyone worked from home while wearing VR headsets. But what they got was an expensive toy that only had niche uses.
AI performs do well on certain coding tasks because a lot of the individual problems that make up a particular piece of software have already been solved. It’s standard practice to design programs as individual units, each of which performs the smallest task possible, and which can then be assembled to complete more complex tasks. This fits very well into the LLM model of assembling pieces into their most likely expected configurations. But it cannot create truly novel code, except by a kind of trial and error mutation process. It cannot problem solve. It cannot identify a users needs and come up with ideal solutions to them. It cannot innovate.
This means that, at best, genAI in the software world becomes a tool for producing individual code elements, guided and shepherded by experienced programmers. It does not replace the software industry, merely augments it, and it does so at a cost that many companies simply may not feel is worth paying.
And that’s its best case scenario. In every other industry AI has been a spectacular failure. But it’s being invested in as if it will be a technological reckoning for every form of intellectual labour on earth. That is the absolute definition of a bubble.
o3 made the high score on ARC through brute force, not by being good. To raise the score from 75% to 87% required 175 times more computing power, but exactly stunning returns.
If it can through brute force, it can do it. That’s the first step towards true agi, nobody said the first AGI would be economical, this feels like a major goalpost shift if you’re acknowledging it can do it at all, isn’t that insane?
A little bit ago, everyone would’ve been saying this will never happen, that there was a natural wall simply because all it does is predict the next token, it’s been like, a few years of llm’s and they’re already getting this insane. We’re going to have AGI soon, it might not be a transformer, but billions upon billions of dollars are being thrown at this problem, there are people smart enough in the world to make this work, and this is the earliest sign that it’s coming.
I’m not convinced that it’s anywhere near an AGI, I’m convinced after combing through papers and code, that it’s an amazing parlor trick.
I’d love to be proven wrong, but everything I’ve seen and everything I’ve used in my studies ( using DNN to simulate neurodivergence and spinal disgenesis, which is kinda AI adjacent) leads me to believe that the current part won’t lead to anything but convincing parlor tricks.
The argument could be made that if a trick is convincing enough, does it matter if it’s intelligent or not.
I’m not entirely sure.
A non-probabilistic algorithm, probably. Something that didn’t rely on the liklihood of association, and instead was capable of context and rationality.
Something that wouldn’t have a system capable of saying “Put glue on your pizza” because it would know that’s a silly thing to say to a human. A system that, when asked "Whats a good caustic detergent " wouldn’t be able to respond "Any good caustic detergent is a good caustic detergent " because duh. Something that doesn’t require thousands of hours of training to update and instead is capable of ingesting and rationalize new information on the fly.
You’re probablistic, you just have an internal verifier, you think things that are silly, and then decide not to say them all the time. A human being often thinks things that they realize are silly before they say them… that’s an entirely unfair goal in the first place from my perspective, why does it have to be non-probablistic?
Are you not a general intelligence because sometimes your brain thinks silly things?
o3 currently works precisely that way, by the way, it generates hundreds of possible things, and then uses something that checks if the steps actually work, before it outputs. In fact, they then reinforce it on these correct logical steps, so it becomes better at not outputting illogical answers like you said.
it’s interesting that you said “not on the probability of the next word, but on context and rationality”
context IS pricesely that, you know what’s likely to come next because of the context, that’s you understanding context. YOU as a human being don’t even always get this right, you must realize we are not perfect beings, we think of possibilities and choose the right one. I think we’re much better at this right now, but i don’t think that’s a fundamental difference between us and o3.
Rationality is the internal verifier.
Something that doesn’t require thousands of hours of training to update and instead is capable of ingesting and rationalize new information on the fly.
Being able to do this is… exactly what arc-agi was testing. Literally the entire point of the benchmark, it can do that.
I’ve done the test by the way, I solved it by brute forcing possible solutions in my head, then checking if they were true… did you just divine the answers instantly?
Where, in that position piece, do they mention o3? Who “proved” this?
Additionally, I’m pretty sure that this “ARC AGI” benchmark is not using the same definition of AGI that you linked to by DeepMind. Conflating them is misleading. There is already so much misinformation out there about “AI”, don’t add to it.
Lastly, I struggle to take at face value essays written by for-profit companies claiming they have AGI (that DeepMind paper links to OpenAI essays). They only stand to gain monetarily by claiming that their AI is an AGI (to be clear, this is an opinion; I do not have evidence to suggest that OpenAI is being disingenuous).
Unless we invent cold fusion between the next 5 years, they will never be economical. They are the most energy inefficient thing ever invented by humanity and all prediction models state that it will cost more energy, not less, to keep making them better. They will never be energy efficient nor economical in their current state, and most companies are out of ideas on how to shake it up. Even the people who created generative models agree that they have just been brute forcing by making the models larger with more energy consumption. When you try to make them smaller or more energy efficient, they fall off the performance cliff and only produce garbage. I’m sure there are researchers doing cool stuff, but it is neither economical nor efficient.
Untrue. There are small models that produce better output than the previous “flagships” like GPT-2. Also, you can achieve much more than we currently do with far less energy by working on novel, specialised hardware (neuromorphic computing).
Why is it getting an AGI stamp now? I was under the impression humanity has not delivered a sentient AI? Which is what the AGI title was supposed to be used for…has that been pulled back again?
I have no idea how people can consider this to be a hype bubble especially after the o3 release. It smashed the ARC AGI benchmark on the performance front. It ranks as the 175th best competitive coder in the world on Codeforces’ leaderboard.
o3 proved that it is possible to have at least an expert AGI if not a Virtuoso AGI (according to Deep mind’s definition of AGI). Sure, it’s not economical yet. But it will get there very soon (just like how the earlier GPTs were a lot dumber and took a lot more energy than the newer, smaller parameter models).
Please remember - fight to seize the means of production. Do not fight the means of production themselves.
It’s a bubble because OpenAI spend $2.35 for every $1.00 they make. Yes, you’re mathing right, that is a net loss.
It’s a bubble because all of the big players in AI development agree that future models will cost exponentially more money to train, for incremental gains. That means there is no path forward that doesn’t intensely amplify the unprofitability of an already deeply unprofitable industry.
It’s a bubble because newer models with better capabilities only cost more and more to run.
It’s a bubble because as far as anyone knows there will never be a solution to the hallucination problem.
It’s a bubble because despite investments treating it as a trillion dollar industry, no one has yet figured out a trillion dollar problem that AI can solve.
You’re trying on a new top of the line VR headset and saying “Wow, this is incredible, how can anyone say this is a bubble?” Its not about how cool the tech is in isolation, it’s about its potential to effect widespread change. Facebook went in hard on VR, imagining a future where everyone worked from home while wearing VR headsets. But what they got was an expensive toy that only had niche uses.
AI performs do well on certain coding tasks because a lot of the individual problems that make up a particular piece of software have already been solved. It’s standard practice to design programs as individual units, each of which performs the smallest task possible, and which can then be assembled to complete more complex tasks. This fits very well into the LLM model of assembling pieces into their most likely expected configurations. But it cannot create truly novel code, except by a kind of trial and error mutation process. It cannot problem solve. It cannot identify a users needs and come up with ideal solutions to them. It cannot innovate.
This means that, at best, genAI in the software world becomes a tool for producing individual code elements, guided and shepherded by experienced programmers. It does not replace the software industry, merely augments it, and it does so at a cost that many companies simply may not feel is worth paying.
And that’s its best case scenario. In every other industry AI has been a spectacular failure. But it’s being invested in as if it will be a technological reckoning for every form of intellectual labour on earth. That is the absolute definition of a bubble.
o3 made the high score on ARC through brute force, not by being good. To raise the score from 75% to 87% required 175 times more computing power, but exactly stunning returns.
Why does it matter?
If it can through brute force, it can do it. That’s the first step towards true agi, nobody said the first AGI would be economical, this feels like a major goalpost shift if you’re acknowledging it can do it at all, isn’t that insane?
A little bit ago, everyone would’ve been saying this will never happen, that there was a natural wall simply because all it does is predict the next token, it’s been like, a few years of llm’s and they’re already getting this insane. We’re going to have AGI soon, it might not be a transformer, but billions upon billions of dollars are being thrown at this problem, there are people smart enough in the world to make this work, and this is the earliest sign that it’s coming.
I’m not convinced that it’s anywhere near an AGI, I’m convinced after combing through papers and code, that it’s an amazing parlor trick.
I’d love to be proven wrong, but everything I’ve seen and everything I’ve used in my studies ( using DNN to simulate neurodivergence and spinal disgenesis, which is kinda AI adjacent) leads me to believe that the current part won’t lead to anything but convincing parlor tricks.
The argument could be made that if a trick is convincing enough, does it matter if it’s intelligent or not.
What would convince you?
I’m not entirely sure.
A non-probabilistic algorithm, probably. Something that didn’t rely on the liklihood of association, and instead was capable of context and rationality.
Something that wouldn’t have a system capable of saying “Put glue on your pizza” because it would know that’s a silly thing to say to a human. A system that, when asked "Whats a good caustic detergent " wouldn’t be able to respond "Any good caustic detergent is a good caustic detergent " because duh. Something that doesn’t require thousands of hours of training to update and instead is capable of ingesting and rationalize new information on the fly.
You’re probablistic, you just have an internal verifier, you think things that are silly, and then decide not to say them all the time. A human being often thinks things that they realize are silly before they say them… that’s an entirely unfair goal in the first place from my perspective, why does it have to be non-probablistic?
Are you not a general intelligence because sometimes your brain thinks silly things?
o3 currently works precisely that way, by the way, it generates hundreds of possible things, and then uses something that checks if the steps actually work, before it outputs. In fact, they then reinforce it on these correct logical steps, so it becomes better at not outputting illogical answers like you said.
it’s interesting that you said “not on the probability of the next word, but on context and rationality”
context IS pricesely that, you know what’s likely to come next because of the context, that’s you understanding context. YOU as a human being don’t even always get this right, you must realize we are not perfect beings, we think of possibilities and choose the right one. I think we’re much better at this right now, but i don’t think that’s a fundamental difference between us and o3.
Rationality is the internal verifier.
Being able to do this is… exactly what arc-agi was testing. Literally the entire point of the benchmark, it can do that.
I’ve done the test by the way, I solved it by brute forcing possible solutions in my head, then checking if they were true… did you just divine the answers instantly?
Where, in that position piece, do they mention o3? Who “proved” this?
Additionally, I’m pretty sure that this “ARC AGI” benchmark is not using the same definition of AGI that you linked to by DeepMind. Conflating them is misleading. There is already so much misinformation out there about “AI”, don’t add to it.
Lastly, I struggle to take at face value essays written by for-profit companies claiming they have AGI (that DeepMind paper links to OpenAI essays). They only stand to gain monetarily by claiming that their AI is an AGI (to be clear, this is an opinion; I do not have evidence to suggest that OpenAI is being disingenuous).
Unless we invent cold fusion between the next 5 years, they will never be economical. They are the most energy inefficient thing ever invented by humanity and all prediction models state that it will cost more energy, not less, to keep making them better. They will never be energy efficient nor economical in their current state, and most companies are out of ideas on how to shake it up. Even the people who created generative models agree that they have just been brute forcing by making the models larger with more energy consumption. When you try to make them smaller or more energy efficient, they fall off the performance cliff and only produce garbage. I’m sure there are researchers doing cool stuff, but it is neither economical nor efficient.
Untrue. There are small models that produce better output than the previous “flagships” like GPT-2. Also, you can achieve much more than we currently do with far less energy by working on novel, specialised hardware (neuromorphic computing).
Your example is strange because, as far as I know, GPTs aren’t economical either.
Why is it getting an AGI stamp now? I was under the impression humanity has not delivered a sentient AI? Which is what the AGI title was supposed to be used for…has that been pulled back again?
Agi has nothing to do with sentience, which cannot be measured, openai, I think validly, defines it as a system that can do all intellectual labor.
So it’s now, can it do anything a human can do?..sans emotional traits.
It was never about sentience, sentience is a meaningless, unmeasurable term.
It’s a question of if it can replace humans in the workforce.
artificial general intelligence means it’s able to generalize its intelligence, not sentience at all.