The underlying problems are still the same. Transformer models do not have an inkling of intelligence, and they still don’t have an internal model of the world around them. Creativity is also non-existent, which is a hard requirement if you want these things to actually create something new instead of rehashing what already exists.
There was no rehashing in 2018. What you did was you trained your own (local) model on some data to help you work. For example, a grad student would painstakingly mark cells in em-scans for weeks and even months, now they could do it for a day, use that result to train a neural network and only supervise the job done. And the job wiuld take minutes instaid of weeks or months.
There were no llms in 2018, IIRC the google/deepmind paper was from 2019 (can be wrong here).
Source: built AI training app in 2018 for science company.
they still don’t have an internal model of the world around them
This has been shown repeatedly to not be correct since 2023.
That’s literally how transformers turned out to have been working so well. They build world models and bags of heuristics fitting the training data to extrapolate from.
It’s almost surprising you managed to use the correct term for the tech but then managed to misrepresent it so egregiously.
You really think that an Othello board could represent the world? Talk about being delusional. This is pathetic. The subheadline: “OTHELLO-GPT USUALLY PREDICTS LEGAL MOVE” says it all - it cannot have a representative internal world model, because in that case the “USUALLY” would go away, as anyone who has played othello for longer than 10 minutes will tell you - and we are talking about a game with very simple rules here, the complexity does not even reach chess - and people expect these things to spit out coherent and valid solutions from target selection in wars to the color their bedroom walls should be painted in.
Edit: I really have to add here the following excerpt (percentages are the rate of invalid moves as the top choice):
The error rate is 0.01% for Othello-GPT trained on the synthetic dataset and 5.17% for Othello-GPT trained on the championship dataset. For comparison, the untrained Othello-GPT has an error rate of 93.29%.
So even if you just feed them real world data without completely synthesizing it, 1 in 20 moves are invalid in a very simple (rule-side) game. If you do not give them a whole lot of played games as a basis, it fails in over 93% of cases (not at making a GOOD move, just at making any valid move!), which is probably very close to a random guess for a move. Something that has even an inkling of an internal world view would be able to adapt after being told that their move is invalid. Also, an internal world view would allow for simply dropping the rulebook onto ChatGPT as a basis, as it works with humans that can imagine.
You really think that an Othello board could represent the world?
No, you seem to be misunderstanding what a “world model” is. That research was on a toy model a fraction of a fraction of the size of modern models and exclusively trained on board notation moves. And yet it modeled the geometry of a board and positions of pieces and ‘my’ vs ‘opponent’ status of those pieces (there’s been multiple follow-ups and replications looking deeper at what encoded).
The question it answered is that transformers don’t simply learn surface statistics from the data but build world models from the data they are trained on.
So when you take a model orders of magnitude larger and train them on collective human culture orders (of orders) of magnitude larger than the board moves in the Othello-GPT paper, what do you think happens? No world modeling at all? Well there’s been a number of interpretability papers since 2023 finding that modem LLMs have functional models of all sorts of things, even including tracking emotions for themselves vs others.
Also, you are assuming that a trained world model is a lossless identical version, which it isn’t and nobody claimed. In fact, arguably the best way to identify AI generated videos these days is by over-expressed emotions of characters. Why? Because the training data is over indexed on influencer exaggerated faces which game clicks so video and image world models of human emotions are miscalibrated.
Transformers have a world model, but often it’s of a kind of parallel imagined/extrapolated world that’s not the real one. Another good example of this is how most models are certain that real world news after their cutoff isn’t true because extrapolating from their internal world model (from the training data before their cutoff) those events seem improbable. The real world doesn’t exactly match the internal model and that drift grows the more they need to extrapolate.
If you want to continue to discuss (I’m happy to help explain and educate you), I’d recommend reading up a bit on what “world model” means in relation to the tech.
I know very well what a “world model’” is, please stop trying to undermine my credibility.
If you feed these things thousands of datasets that only consist of a specific set of coordinates, your output will also only consist of these coordinates, which will create the illusion that they actually have knowledge about how an game board looks (as i said, these things are remixers after all). And feeding it a championship data set, where not one invalid move will appear, should ensure at least a close to 100% of valid moves if they were actually able to have an actual internal model of the system they are trying to manipulate. (and in reality, it should choose pretty good moves too, but this was not evaluated at all…).
Their test cases (and the error rates!) were not even “play a game of Othello”, but feeding them gamestates and asking them for a legal move. IMO, this whole setup does not require any form or shape of a world model, and if it actually existed, the error rates would be a ton lower - the championship data will not contain a single invalid move, and the concept of choosing an empty space on the board, touching the other players piece with one of my own ones on the other side of the mass of pieces, is really not a hard one.
I agree that current models will probably have a better performance, but this is NOT because of a better “world model”, but because the statistical sieve has become a lot finer by blowing up the amount of parameters.
I use Claude every day for actual production software used in a real product and I can say with confidence that even the most advanced publicly available models totally lack the higher order thinking required to build good software. They’re pretty good at writing code, when directed well. They are not good at architectural or design decisions and they’re even worse at actually following decisions made outside of their context. Those are the actually challenging parts of software engineering and you have to beat them into the model every step of the way or your software quickly becomes an unmaintainable mess.
Then someone has write the spec and develop the verification. Which is, again, the hard part. If you’ve ever worked when a business or requirements team, you’ll also know that the spec and verification are constantly moving targets. With a lot of time and attention, you can create a whole ecosystem of agents that will output somewhat acceptable code at scale (or so I’ve heard from people I trust), but that just moves the human effort required to develop and maintain the software over to developing and maintaining the automation. Then if anything breaks catastrophically in a way AI can’t fix, your team’s actual software skills have totally stagnated.
Most human programmers are bad at architecture. They’re making lots of mistakes too. Perfection is not a criterion for intelligence. In that case no human is intelligent.
Yeah, that’s why software engineering is a valuable skill that demands significant compensation. Even many of the people who do it for a living aren’t that great at it. I don’t even consider myself to be an especially skilled software engineer. There is a lot more to the job than simply writing code and it takes several years to become truly competent, with lots of mistakes along the way.
Go ahead and read my comment again and let me know where I demand perfection or even talk about the criteria for intelligence. Defining intelligence is (imo) more of a philosophical question than a practical one so I don’t find a lot of practical value in speculating on whether or not LLMs posses some kind of intelligence. It’s not really relevant to what I’m talking about.
My point is that, as it stands right now, I use these models on a daily basis to complete actual work and they cannot be trusted to competently make the kinds of decisions required to build good, maintainable software. Sometimes, if you’re working on a well known common design structure or a common problem, an LLM can pantomime good design decisions (because it’s well represented in the data set) but as soon as you step into a situation that requires actual creativity or must account for non-obvious pitfalls in the codebase, it’s a crapshoot. Far worse than an inexperienced software engineer. I believe this is an inherent limitation of transformer based models because they seem to lack any kind of mechanism for this type of higher order thinking.
My comment was: AI in 2018 is intelligent in a very different way from AI in 2026.
The common thread I see is: well AI can do these things well, but fail at these other things. Therefore, it cannot be intelligent.
Even if the things it fails at are things humans would normally fail at. Any test for intelligence in AI must also apply for humans!
I want to have a clear cut answer: why isn’t AI of 2026 considered intelligent. No one in this entire thread has given a clear answer - besides gut feelings because they feel uncomfortable with the idea that AI is intelligent.
Sure, there’s been a lot of progress in the field and AI models are certainly more useful now than they were in 2018, but I think that’s mostly do to 1) the size and scale of the models, and 2) the harnesses around them. I think that commenter is right in that there are some apparently inherent limitations to the technology (by which I mean transformer based models) which still persist today.
well AI can do these things well, but fail at these other things. Therefore, it cannot be intelligent.
Well that’s a dumb argument and one I’m not making. After using these things quite a bit for actual work, it’s clear to me personally that these things are pretty good at parsing data and spitting out statistically correct responses, based on the training data, but that there is no real reasoning or creativity happening in the process.
Even if the things it fails at are things humans would normally fail at. Any test for intelligence in AI must also apply for humans!
Why? Why would our standards for people be the same as our standards for a machine? There are many things that an LLM is much better at than any human will ever be (spitting out large amounts of mostly coherent text in a short amount of time, searching and parsing large amounts of text in a short amount of time, etc) and some things that an LLM is way worse at than most people (counting the number of letters in a word when that specific scenario hasn’t be well covered in training, due to how tokenization works. Arithmetic, etc.) There are many many reasons why people and machines are not really comparable, I don’t think inventing a machine that’s pretty good at natural language changes that.
I want to have a clear cut answer: why isn’t AI of 2026 considered intelligent. No one in this entire thread has given a clear answer - besides gut feelings because they feel uncomfortable with the idea that AI is intelligent.
I think the reason you won’t get a clear cut answer is that there isn’t really a clear cut definition of what intelligence even is or how to quantify it in any objective way. If you ask 20 people “what is intelligence?” or how to tell if something possesses it, you’ll get 20 different answers. Our own methods of rating our own intelligence are notoriously subjective. People who have devoted their entire professional life to this subject will tell you how little we know. I am not uncomfortable with the idea that AI is intelligent because I know that it isn’t. I can tell you why I feel that way but this comment is long enough and I have a feeling it won’t satisfy you anyway.
All I’ll say is it’s pretty irritating to hear a bunch of people insist that this thing is going to take over your job and then world, and then when you actually use it, you quickly see all the gaping holes in its capabilities, despite the impressive first impression. It becomes pretty clear pretty quickly that this is a useful new technology but not exactly the next industrial revolution. At least not as it currently exists. I think we’ve already started to hit diminishing returns, but maybe I’m wrong. I can’t help but feel like the people still smugly replying to “AI kinda sucks actually” with “well you must not be using it right 😏” at this point either don’t know, don’t care, or are actively lying for financial or ideological reasons.
Intelligence isn’t a one-dimensional axis either. Intelligence exists in many facets. Agents (so LLMs combined with tools, memory, etc.) overlap pretty much with one aspect that we perceive as intelligent. The main risk is basically the same as with humans of similar one-sided intellect: if you expect them to be equally intelligent in other areas, you will get disappointed or even hurt.
I am not a non-user. I run local models, because i am a tech enthusiast, and am watching the space closely. I will never spend a cent or use the offerings of OpenAI and Anthropic tho - i will not finance the people who are providing the tools to bind us all.
Open weight models are not far behind and good enough to get software developers jobless, but there is a huge difference in what you can do with 1M context and 500token/sec or 60k context and 40token/sec.
Sure! End the AI Bubble and i’ll go ahead and buy a bunch of H200’s.
But for real, if you structure your tasks well, you can get by for most things with local models, and for a small business in IT, investing in dedicated hardware you own is a smart move. The employer of an acquaintance of mine bought some inference hardware for his 5 engineer business, and they rely mostly on that. Around 80% of all tasks are done by a GLM5.3 model, and while they currently still use Claude for some things, the percentage of external inference has dropped dramatically in the last months, because people have adapted on how to structure tasks for the local model.
Noone can explain to me how Claude, OpenAI and the NeoClouds plan to become profitable businesses when you can reduce inference costs by 90% by getting your own inference - and it’s not even that hard to set up. if you already have IT staff for your inhouse stuff, they will probably manage.
Someone will come up with a decent accelerator eventually. Right now we’re in the pre-soundblaster days where audio sythesis is being done by the CPU but for neural networks.
We’re repurposing graphics hardware, these are primitive times.
I have to disagree here. We are on the absolute edge of current physics regarding miniaturization - without a massive breakthrough in quantum physics on how we can use subatomic particles for transistors we are nearing the maximum we can shrink these circuits. These are already optimized for exactly this kind of math - a Vera Rubin will never be able to render a game. Also, the mathematical basis (a ton of matrix multiplication) does not give much hope that we can replace all this math with something simpler. Maybe a radical redesign taking a completely different approach can help, but i wouldn’t count on that - it would mean reinventing the whole damn thing on a completely new base.
I can’t really see any explanation when i take the above into account.
This is not a “new approach”, it’s the same one but hardware-based. It also is wildly impractical. Models are only useful if they stay somewhat up to date. This way only produces a whole load of garbage everytime you want to update it. Probably useful in the industry or robotics, but not for an everyday LLM.
In any case, confidence without knowledge about the matter, that is something we laughed at LLMs just 2 years ago. Those became much much better at it.
LLMs have made a number of significant mathematical discoveries in the past year. Some of them were problems that had significant prize money attached and were considered some of the biggest open problems in mathematics. This has been so disruptive that Terrance Tao, who is widely considered to be the worlds greatest living mathematician, has spent considerable time publicly grappling with how the field of mathematics should proceed now that AI is in the picture (for example, he thinks less importance should be placed on discovery, since that is now trivial, and more emphasis should be placed on understanding).
I am sympathetic to criticisms of LLMs, but these criticisms need to be grounded in reality. We can’t just put our head in the sand. To say that AI has “not an inkling of intelligence” is just factually inaccurate at this point.
You do not simply steal groundbreaking mathematical theorems. These are open problems that haven’t been solved, so prior to these AI proofs no solution had been discovered, so there was nothing there to steal.
Dunno about that. Cause they did steal them. Assuming their solutions are even correct, cause nobody has done peer review on them, because LLMs can’t publish and anthropic or whoever refuses to engage with the academic community.
All this and there’s a lot of talk about stealing results among maths academics right now. Including a lack of attribution.
Hard to say exactly. I know my comments have been scraped by ChatGPT and other AI companies. But I don’t know how much of what I’ve said has contributed.
Also, when they have such funding, they can pay me off more than an award or the fame would be worth.
Are you suggesting that these open problems in mathematics were in fact solved by someone who did not take any prize money for them or claim any credit, unbeknownst to the entire mathematical community, and the AI stole these secret solutions? Because in order for these solutions to be stolen that would need to be the scenario.
I am suggesting that someone was in the process of finalising publication of papers which might have asserted these. Papers such as these can take more than a year as a draft paper.
Draft papers are often made public for gasppeer review
There have been accusations by mathematicians that OpenAI used their (non-public, but fed to OpenAIs LLM) methods. It would have been very possible to take what they already did and throw as much compute as they had available at it as soon as the hard parts were done to race ahead at the final stretch.
LLMs fail badly at trying to do original research. There was a study where 4 not-yet published papers were used - they gave the LLM all the data the researchers had at the beginning of the study. The LLMs were not successful at reproducing anything of it.
These things are not intelligent. Just because they can remix existing content does not mean they have creativity. And they miss the entire “physicality”-aspect of human existence, which is a central part of how intelligence develops - by interacting with the world and creating an internal model of how it works. They do not have a sense of self - a bunch of matrix multiplication will never achieve this.
They can’t be creative, because for creativity (and therefor intelligence - these two things go hand in hand) you must be able to have thoughts that have never been thought in a specific context. All these transformer models do is to remix existing thoughts - this looks a bit like intelligence, but the contained data can never be expanded outside of training a new model at a large expense. Intelligence also means to be able to learn and adapt - outside of creating a text file where LLMs can store their non-original output and reading it again, they do not have this capability, especially the “adaptation” part.
Do not humanize these mathematical constructs. This thinking is what got us into the current mess to begin with.
I personally like the technology aspect of Transformer Models a lot! But what is done with it is not in the interest of humanity - at all. Just this morning we all were able to read in the news that Anthropic uses its LLM to do a “Minority Report” and contact the police when their LLM, which is being used to scrape socials and categorize people depending on how much “Threat” they pose, tells them someone might do something illegal in the future.
This is the real usage of this technology. We are nearing a future where our every step, action, word, and even thoughts in some cases (through the wonders of profiling) are ingested by one of these things, and we will get sorted into bins according to threat level and how much money we can provide them with. Tip your toe over the line by saying that this sucks ass, and find yourself in your friendly neighbourhood reeducation gulag.
If it was only Navier-Stokes I would agree. Now it has produced significant results for hundreds of high profile problems - many of which would be Fields medal worthy if solved by a human. It’s vanishingly unlikely that all of these already had unpublished solutions waiting.
I’m not trying to say that LLMs are on the whole good for humanity. I agree with you that they’re not. And I’m not trying to argue that OpenAI never steals intellectual property. Clearly they do.
But these mathematical problems were solved by OpenAI’s models. That much is indisputable. There was nothing in the training data that could have given them the solutions.
If you feed them the approach the mathematicians were using, they for sure can solve it! It has never been published what source material their LLM was provided with. This is the equivalent of someone telling you a result, but never telling you how they got the information needed to achieve it in the first place.
There is value in this technology. But we should do out best to not conflate remixing shit with intelligence - these things are not the same at all. LLMs are more like the proverbial infinite room of monkeys with typewriters, with the very important restriction that these monkeys can never type something that resembles an original thought.
The underlying problems are still the same. Transformer models do not have an inkling of intelligence, and they still don’t have an internal model of the world around them. Creativity is also non-existent, which is a hard requirement if you want these things to actually create something new instead of rehashing what already exists.
There was no rehashing in 2018. What you did was you trained your own (local) model on some data to help you work. For example, a grad student would painstakingly mark cells in em-scans for weeks and even months, now they could do it for a day, use that result to train a neural network and only supervise the job done. And the job wiuld take minutes instaid of weeks or months.
There were no llms in 2018, IIRC the google/deepmind paper was from 2019 (can be wrong here).
Source: built AI training app in 2018 for science company.
This has been shown repeatedly to not be correct since 2023.
That’s literally how transformers turned out to have been working so well. They build world models and bags of heuristics fitting the training data to extrapolate from.
It’s almost surprising you managed to use the correct term for the tech but then managed to misrepresent it so egregiously.
You really think that an Othello board could represent the world? Talk about being delusional. This is pathetic. The subheadline: “OTHELLO-GPT USUALLY PREDICTS LEGAL MOVE” says it all - it cannot have a representative internal world model, because in that case the “USUALLY” would go away, as anyone who has played othello for longer than 10 minutes will tell you - and we are talking about a game with very simple rules here, the complexity does not even reach chess - and people expect these things to spit out coherent and valid solutions from target selection in wars to the color their bedroom walls should be painted in.
Edit: I really have to add here the following excerpt (percentages are the rate of invalid moves as the top choice):
So even if you just feed them real world data without completely synthesizing it, 1 in 20 moves are invalid in a very simple (rule-side) game. If you do not give them a whole lot of played games as a basis, it fails in over 93% of cases (not at making a GOOD move, just at making any valid move!), which is probably very close to a random guess for a move. Something that has even an inkling of an internal world view would be able to adapt after being told that their move is invalid. Also, an internal world view would allow for simply dropping the rulebook onto ChatGPT as a basis, as it works with humans that can imagine.
No, you seem to be misunderstanding what a “world model” is. That research was on a toy model a fraction of a fraction of the size of modern models and exclusively trained on board notation moves. And yet it modeled the geometry of a board and positions of pieces and ‘my’ vs ‘opponent’ status of those pieces (there’s been multiple follow-ups and replications looking deeper at what encoded).
The question it answered is that transformers don’t simply learn surface statistics from the data but build world models from the data they are trained on.
So when you take a model orders of magnitude larger and train them on collective human culture orders (of orders) of magnitude larger than the board moves in the Othello-GPT paper, what do you think happens? No world modeling at all? Well there’s been a number of interpretability papers since 2023 finding that modem LLMs have functional models of all sorts of things, even including tracking emotions for themselves vs others.
Also, you are assuming that a trained world model is a lossless identical version, which it isn’t and nobody claimed. In fact, arguably the best way to identify AI generated videos these days is by over-expressed emotions of characters. Why? Because the training data is over indexed on influencer exaggerated faces which game clicks so video and image world models of human emotions are miscalibrated.
Transformers have a world model, but often it’s of a kind of parallel imagined/extrapolated world that’s not the real one. Another good example of this is how most models are certain that real world news after their cutoff isn’t true because extrapolating from their internal world model (from the training data before their cutoff) those events seem improbable. The real world doesn’t exactly match the internal model and that drift grows the more they need to extrapolate.
If you want to continue to discuss (I’m happy to help explain and educate you), I’d recommend reading up a bit on what “world model” means in relation to the tech.
I know very well what a “world model’” is, please stop trying to undermine my credibility.
If you feed these things thousands of datasets that only consist of a specific set of coordinates, your output will also only consist of these coordinates, which will create the illusion that they actually have knowledge about how an game board looks (as i said, these things are remixers after all). And feeding it a championship data set, where not one invalid move will appear, should ensure at least a close to 100% of valid moves if they were actually able to have an actual internal model of the system they are trying to manipulate. (and in reality, it should choose pretty good moves too, but this was not evaluated at all…).
Their test cases (and the error rates!) were not even “play a game of Othello”, but feeding them gamestates and asking them for a legal move. IMO, this whole setup does not require any form or shape of a world model, and if it actually existed, the error rates would be a ton lower - the championship data will not contain a single invalid move, and the concept of choosing an empty space on the board, touching the other players piece with one of my own ones on the other side of the mass of pieces, is really not a hard one.
I agree that current models will probably have a better performance, but this is NOT because of a better “world model”, but because the statistical sieve has become a lot finer by blowing up the amount of parameters.
Spoken with the confidence of a non user.
I use Claude every day for actual production software used in a real product and I can say with confidence that even the most advanced publicly available models totally lack the higher order thinking required to build good software. They’re pretty good at writing code, when directed well. They are not good at architectural or design decisions and they’re even worse at actually following decisions made outside of their context. Those are the actually challenging parts of software engineering and you have to beat them into the model every step of the way or your software quickly becomes an unmaintainable mess.
Use Spec driven development and a good verification loop to overcome the “beat it into the model at every step”.
Then someone has write the spec and develop the verification. Which is, again, the hard part. If you’ve ever worked when a business or requirements team, you’ll also know that the spec and verification are constantly moving targets. With a lot of time and attention, you can create a whole ecosystem of agents that will output somewhat acceptable code at scale (or so I’ve heard from people I trust), but that just moves the human effort required to develop and maintain the software over to developing and maintaining the automation. Then if anything breaks catastrophically in a way AI can’t fix, your team’s actual software skills have totally stagnated.
Most human programmers are bad at architecture. They’re making lots of mistakes too. Perfection is not a criterion for intelligence. In that case no human is intelligent.
Yeah, that’s why software engineering is a valuable skill that demands significant compensation. Even many of the people who do it for a living aren’t that great at it. I don’t even consider myself to be an especially skilled software engineer. There is a lot more to the job than simply writing code and it takes several years to become truly competent, with lots of mistakes along the way.
Go ahead and read my comment again and let me know where I demand perfection or even talk about the criteria for intelligence. Defining intelligence is (imo) more of a philosophical question than a practical one so I don’t find a lot of practical value in speculating on whether or not LLMs posses some kind of intelligence. It’s not really relevant to what I’m talking about.
My point is that, as it stands right now, I use these models on a daily basis to complete actual work and they cannot be trusted to competently make the kinds of decisions required to build good, maintainable software. Sometimes, if you’re working on a well known common design structure or a common problem, an LLM can pantomime good design decisions (because it’s well represented in the data set) but as soon as you step into a situation that requires actual creativity or must account for non-obvious pitfalls in the codebase, it’s a crapshoot. Far worse than an inexperienced software engineer. I believe this is an inherent limitation of transformer based models because they seem to lack any kind of mechanism for this type of higher order thinking.
My comment was: AI in 2018 is intelligent in a very different way from AI in 2026.
The common thread I see is: well AI can do these things well, but fail at these other things. Therefore, it cannot be intelligent.
Even if the things it fails at are things humans would normally fail at. Any test for intelligence in AI must also apply for humans!
I want to have a clear cut answer: why isn’t AI of 2026 considered intelligent. No one in this entire thread has given a clear answer - besides gut feelings because they feel uncomfortable with the idea that AI is intelligent.
Sure, there’s been a lot of progress in the field and AI models are certainly more useful now than they were in 2018, but I think that’s mostly do to 1) the size and scale of the models, and 2) the harnesses around them. I think that commenter is right in that there are some apparently inherent limitations to the technology (by which I mean transformer based models) which still persist today.
Well that’s a dumb argument and one I’m not making. After using these things quite a bit for actual work, it’s clear to me personally that these things are pretty good at parsing data and spitting out statistically correct responses, based on the training data, but that there is no real reasoning or creativity happening in the process.
Why? Why would our standards for people be the same as our standards for a machine? There are many things that an LLM is much better at than any human will ever be (spitting out large amounts of mostly coherent text in a short amount of time, searching and parsing large amounts of text in a short amount of time, etc) and some things that an LLM is way worse at than most people (counting the number of letters in a word when that specific scenario hasn’t be well covered in training, due to how tokenization works. Arithmetic, etc.) There are many many reasons why people and machines are not really comparable, I don’t think inventing a machine that’s pretty good at natural language changes that.
I think the reason you won’t get a clear cut answer is that there isn’t really a clear cut definition of what intelligence even is or how to quantify it in any objective way. If you ask 20 people “what is intelligence?” or how to tell if something possesses it, you’ll get 20 different answers. Our own methods of rating our own intelligence are notoriously subjective. People who have devoted their entire professional life to this subject will tell you how little we know. I am not uncomfortable with the idea that AI is intelligent because I know that it isn’t. I can tell you why I feel that way but this comment is long enough and I have a feeling it won’t satisfy you anyway.
All I’ll say is it’s pretty irritating to hear a bunch of people insist that this thing is going to take over your job and then world, and then when you actually use it, you quickly see all the gaping holes in its capabilities, despite the impressive first impression. It becomes pretty clear pretty quickly that this is a useful new technology but not exactly the next industrial revolution. At least not as it currently exists. I think we’ve already started to hit diminishing returns, but maybe I’m wrong. I can’t help but feel like the people still smugly replying to “AI kinda sucks actually” with “well you must not be using it right 😏” at this point either don’t know, don’t care, or are actively lying for financial or ideological reasons.
Intelligence isn’t a one-dimensional axis either. Intelligence exists in many facets. Agents (so LLMs combined with tools, memory, etc.) overlap pretty much with one aspect that we perceive as intelligent. The main risk is basically the same as with humans of similar one-sided intellect: if you expect them to be equally intelligent in other areas, you will get disappointed or even hurt.
I am not a non-user. I run local models, because i am a tech enthusiast, and am watching the space closely. I will never spend a cent or use the offerings of OpenAI and Anthropic tho - i will not finance the people who are providing the tools to bind us all.
Open weight models are not far behind and good enough to get software developers jobless, but there is a huge difference in what you can do with 1M context and 500token/sec or 60k context and 40token/sec.
Sure! End the AI Bubble and i’ll go ahead and buy a bunch of H200’s.
But for real, if you structure your tasks well, you can get by for most things with local models, and for a small business in IT, investing in dedicated hardware you own is a smart move. The employer of an acquaintance of mine bought some inference hardware for his 5 engineer business, and they rely mostly on that. Around 80% of all tasks are done by a GLM5.3 model, and while they currently still use Claude for some things, the percentage of external inference has dropped dramatically in the last months, because people have adapted on how to structure tasks for the local model.
Noone can explain to me how Claude, OpenAI and the NeoClouds plan to become profitable businesses when you can reduce inference costs by 90% by getting your own inference - and it’s not even that hard to set up. if you already have IT staff for your inhouse stuff, they will probably manage.
Someone will come up with a decent accelerator eventually. Right now we’re in the pre-soundblaster days where audio sythesis is being done by the CPU but for neural networks.
We’re repurposing graphics hardware, these are primitive times.
Also, there, I explained it for you.
I have to disagree here. We are on the absolute edge of current physics regarding miniaturization - without a massive breakthrough in quantum physics on how we can use subatomic particles for transistors we are nearing the maximum we can shrink these circuits. These are already optimized for exactly this kind of math - a Vera Rubin will never be able to render a game. Also, the mathematical basis (a ton of matrix multiplication) does not give much hope that we can replace all this math with something simpler. Maybe a radical redesign taking a completely different approach can help, but i wouldn’t count on that - it would mean reinventing the whole damn thing on a completely new base.
I can’t really see any explanation when i take the above into account.
It was already done. https://chatjimmy.ai/
This is not a “new approach”, it’s the same one but hardware-based. It also is wildly impractical. Models are only useful if they stay somewhat up to date. This way only produces a whole load of garbage everytime you want to update it. Probably useful in the industry or robotics, but not for an everyday LLM.
In any case, confidence without knowledge about the matter, that is something we laughed at LLMs just 2 years ago. Those became much much better at it.
Now suddenly it’s en vogue with humans.
Hey that’s a good zinger, good work. I found an article to help you take down that AI Hater. If you want me to jazz it up, or reword future zingers for better impact, just let me know. I’m down for anything.
LLMs have made a number of significant mathematical discoveries in the past year. Some of them were problems that had significant prize money attached and were considered some of the biggest open problems in mathematics. This has been so disruptive that Terrance Tao, who is widely considered to be the worlds greatest living mathematician, has spent considerable time publicly grappling with how the field of mathematics should proceed now that AI is in the picture (for example, he thinks less importance should be placed on discovery, since that is now trivial, and more emphasis should be placed on understanding).
I am sympathetic to criticisms of LLMs, but these criticisms need to be grounded in reality. We can’t just put our head in the sand. To say that AI has “not an inkling of intelligence” is just factually inaccurate at this point.
And we will never learn how many of these discoveries are born out of simple theft.
Statistics do not have intelligence.
You do not simply steal groundbreaking mathematical theorems. These are open problems that haven’t been solved, so prior to these AI proofs no solution had been discovered, so there was nothing there to steal.
Dunno about that. Cause they did steal them. Assuming their solutions are even correct, cause nobody has done peer review on them, because LLMs can’t publish and anthropic or whoever refuses to engage with the academic community.
All this and there’s a lot of talk about stealing results among maths academics right now. Including a lack of attribution.
So yeah. They probably did steal that shit
I doubt they stole the results of quasi Riemann hypothesis, unique games conjecture and exponential time hypothesis.
Doubt all you like. But the idea a company stole something is hardly controversial or unlikely.
Only if that thing exists to steal in the first place. If it was stolen the people working on it would have made a claim, no?
Hard to say exactly. I know my comments have been scraped by ChatGPT and other AI companies. But I don’t know how much of what I’ve said has contributed.
Also, when they have such funding, they can pay me off more than an award or the fame would be worth.
Are you suggesting that these open problems in mathematics were in fact solved by someone who did not take any prize money for them or claim any credit, unbeknownst to the entire mathematical community, and the AI stole these secret solutions? Because in order for these solutions to be stolen that would need to be the scenario.
I am suggesting that someone was in the process of finalising publication of papers which might have asserted these. Papers such as these can take more than a year as a draft paper.
Draft papers are often made public for gasp peer review
And they robbed that person.
Yes
There have been accusations by mathematicians that OpenAI used their (non-public, but fed to OpenAIs LLM) methods. It would have been very possible to take what they already did and throw as much compute as they had available at it as soon as the hard parts were done to race ahead at the final stretch.
LLMs fail badly at trying to do original research. There was a study where 4 not-yet published papers were used - they gave the LLM all the data the researchers had at the beginning of the study. The LLMs were not successful at reproducing anything of it.
These things are not intelligent. Just because they can remix existing content does not mean they have creativity. And they miss the entire “physicality”-aspect of human existence, which is a central part of how intelligence develops - by interacting with the world and creating an internal model of how it works. They do not have a sense of self - a bunch of matrix multiplication will never achieve this.
They can’t be creative, because for creativity (and therefor intelligence - these two things go hand in hand) you must be able to have thoughts that have never been thought in a specific context. All these transformer models do is to remix existing thoughts - this looks a bit like intelligence, but the contained data can never be expanded outside of training a new model at a large expense. Intelligence also means to be able to learn and adapt - outside of creating a text file where LLMs can store their non-original output and reading it again, they do not have this capability, especially the “adaptation” part.
Do not humanize these mathematical constructs. This thinking is what got us into the current mess to begin with.
I personally like the technology aspect of Transformer Models a lot! But what is done with it is not in the interest of humanity - at all. Just this morning we all were able to read in the news that Anthropic uses its LLM to do a “Minority Report” and contact the police when their LLM, which is being used to scrape socials and categorize people depending on how much “Threat” they pose, tells them someone might do something illegal in the future.
This is the real usage of this technology. We are nearing a future where our every step, action, word, and even thoughts in some cases (through the wonders of profiling) are ingested by one of these things, and we will get sorted into bins according to threat level and how much money we can provide them with. Tip your toe over the line by saying that this sucks ass, and find yourself in your friendly neighbourhood reeducation gulag.
If it was only Navier-Stokes I would agree. Now it has produced significant results for hundreds of high profile problems - many of which would be Fields medal worthy if solved by a human. It’s vanishingly unlikely that all of these already had unpublished solutions waiting.
I’m not trying to say that LLMs are on the whole good for humanity. I agree with you that they’re not. And I’m not trying to argue that OpenAI never steals intellectual property. Clearly they do.
But these mathematical problems were solved by OpenAI’s models. That much is indisputable. There was nothing in the training data that could have given them the solutions.
If you feed them the approach the mathematicians were using, they for sure can solve it! It has never been published what source material their LLM was provided with. This is the equivalent of someone telling you a result, but never telling you how they got the information needed to achieve it in the first place.
There is value in this technology. But we should do out best to not conflate remixing shit with intelligence - these things are not the same at all. LLMs are more like the proverbial infinite room of monkeys with typewriters, with the very important restriction that these monkeys can never type something that resembles an original thought.
Removed by mod
Here are some of the papers, these are a little old.
https://github.com/openai/math/tree/main/
Not our fault if you’re too stupid/uneducated to even read the titles, let alone the accompanying proofs.
No.
Come on now. You can hate AI like most of us on lemmy but if you do not align your views with reality you will gradually slide into delusion.