At what point would you say "yes, this is AGI"? To find out more about the ASI Alliance and why we (as in humanity) needs them, check out the video over here czcams.com/video/2kwfJosYJdE/video.html
It isn't evidence that the technology is plateauing. Its basically a minor update on the current model. Lets wait for gpt5 before we go throwing big words around, shall we?
We've seen the plateauing cross different LLM's. Also these are not my words but the words of different scientific articles and papers I've made videos about in the past. Also, how about we just talk and discuss these things properly instead of trying to tell each other what we can or can not say, shall we? ;)
@@WieseTechnology im mostly using it for finance python code. id say you need to point it in the right direction and a lot of the time paste in the right documentation but after that it makes things pretty simple also id say gpt's best feature is that it has very shallow depth in a massive amount of subjects so ideally string unconnected concepts together and then combining concepts in new ways is where it really shines. like for me combining exponential HHPS request time outs to algo trading to avoid when the stock market has unfavorable positions to avoid losses would be one example. ie if i make a trade and it was a loss ignore any incoming buy signals for 5 days if i make 2 losing trades ignore incoming buy signals for 15 days stuff like that where they mimic other ideas found in HTTP requests.
i know this technology is being said as being some sort of leap forward. What I have noticed in playing around with this, is that this "latency" thing isn't really a thing. One way a human delays answering a question and provide themselves with a few seconds to formulate an answer is by repeating the question. ChatGPT does this. Try it yourself. All that time responding to you and reading back what you asked is simply giving it more time to formulate the answer to the real question. The tech is cool, but this focus on the latency .. . well, not sure it's all that it's being hyped up for.
Yeah I agree, you can also notice how GPT uses fillers to have more time to come up with an actual answer. In my opinion latency is also not a big deal. Who really cares if the AI answers in 1 second or in 0.3 seconds. Not really anything else but a marketing gimmick
It will probably roll out over the next weeks for everyone, starting with America. But not all functions will roll out. For now, I think only the improved text capabilities will be available with hopefully the voice/video capabilities coming soon after
AGI (Artifical GENERAL Intelligence|) is not to be mistaken with ASI (Artifical Super Intelligence|). You fail to understand this tool is beating top expert level humans in any test we can come up with. Consider theory of mind or other causal reasoning, you think average or median humans even consider deep thought that is beyond intuition. Average performance has been achieved a while ago, they are just patching up the hole with great performance in other modals.
Yes obviously AGI isn't to be mistaken with ASI. Still there are many different definitions for AGI and even more definitions on what "intelligence" actually means. LLMs have great knowledge because they are trained on vast amounts of data. Yet if you dig deeper in certain areas they fail to understand things or reason properly beyond a surface level understanding. While it might seem they are "intelligent" mostly they are "knowledgeable" which are two very different things. For example, no one who wants to be taken serious would claim we achieved AGI because an AI could beat the best human in Go. Aside from that I also haven't heard from ANY noteworthy AI researcher that AGI has been achieved. I really appreciate your input and would love to talk further but I hope you, just as me, understand that you probably don't understand much when it comes to this level of AI research and leave polemic claims and assumptions like "You fail to understand" out of any further conversation :)
@@WieseTechnology My point was that average humans don't have beyond surface level understanding on most things. You are again comparing to things expert level humans have. Which is not the point to AGI, but ASI. And ironically, you mentioned that we have only surface level knowledge in the level of AI research.
@@Quaintcy My point was that neither you or me understand AI/AGI on a level that Yann LeCun, Ilya Sutskever, Ben Goertzel, or Sam Altmann understand it. And neither one of them or other AI experts of the same kaliber have claimed that we have achieved AGI yet. So it's interesting that you apparently claim to understand it better than them when saying that we have (sorry if this comes across as a bit polemic from my side but thats how I understand your statement). And it also seems to me that you have a very different definition of AGI than I have or what most people would think AGI means. We have an intelligent computer system that is very knowledgeable, easily more knowledgeable than most humans in most fields. But that has not much to do with actual AGI. By that definition, you could say a very big encyclopedia with vast amounts of knowledge would be an AGI. These AI systems don't have a sense of self, no intrinsic goals, no sophisticated theory of mind, not an actual world model or understanding of it, and lack many other things that one would think an intelligent being should posses. ASI is also no knowledge of what expert level humans have but FAT beyond. Let's try to use IQ to understand ASI here, even tho of course it's lacking many other fundamental things that would need to be in place. But just as an example to understand ASI. An average human has an IQ of 100. An AGI should then also have an IQ of roughly 100 and an ASI system would have an IQ of 500, 1000, 5000, and higher. For an actual developed ASI we humans would in comparison be basically ants.
@@Quaintcy By your definition, if I just made something that recites the wikipedia article on any topic I ask about, it would be smarter than most people because most people don't know that much about most things. It has nothing to do with knowledge, or the ability to answer questions in a test. It is about cognitive ability, very different thing. AGI requires the ability to use deductive, inductive and abductive reasoning. It requires a world model. Machine learning is about induction, and then manually patched with deduction when that fails. Without real abduction, it is impossible for AI to really learn or understand anything, or navigate the real world. No matter how much data you train it on, you cannot send a robot to wander about in a city and run a few errands, because anything that happens that doesn't closely resemble something in its training data or programming will utterly confuse it. That's the difference. There is no broad set of cognitive tasks that an AI can perform as well or better than humans, even if it can recite memorised answers to test questions.
@@WieseTechnology Right, to be clear, my definition of AGI revolves around the term General. You can either define it median, mode or 75 percentile performance in all the test of intelligence we can come up with and ones we haven't, that include metrics for sense of self, theory of mind, world model. As we define intelligence by those measures, it is the only definition of intelligence we know of, it is intelligence as humans can tell. Surpassing 75 percentile would be AGI, but we have not have developed those metrics. AGI is not official. Average IQ is only 100 for a reason. IQ tests spatial reasoning, which can be argued is part of world model. We've not set the bar very high for AI. I did find a article claim gpt4 score 85 on the mensa test and claud 3 got a 101. No result on gpt 4o yet. 4o is not behind if claude 3 can do it. I don't even know what sense of self or theory of mind would test. I do not believe there is consciousness or free will. Which is another whole discussion. Those metric therefore is not included in my definition. I'm not even going to respond to the appeal to authority fallacy. More interesting if you quote them. I suspect their reluctance is more likely due to polics of such statement. Then the other half of expert can't even define AGI, or uses metaphysics. Then they rage about the quailia is not right, as if it's something that can be nailed down, or is even the same between people. They argue based on a feeling.
So you think that GPT4-o is AGI? Interesting, maybe you care to elaborate on your point and thoughts? Ideally without insults or polemic language and instead with actual arguments and well founded knowledge. Should be especially Interesting since not even anyone from OpenAI or any reputable AI scientist would claim that GTP4-o would constitute as AGI aside from one guy on X. But very interested in your point of view. Having an actual conversation is usually something I prefer :)
@@WieseTechnology Actually if you are so smart, dont film your edits with a mobile phone ^_^ Maybe fix the videos sound and maybe try to talk a little less histeric.... i had to silence you for a splitsecond because you were fucking screaming, compared to every other youtube video i watched before Und wenn das alles hinbekommen hast, dann kannste noch an deinem reudigen englisch übenAsk chat gpt how to do these things.....
@@darthfather1294 Look this is exactly what I mean. I never claimed I'm smart and I'm definitely far from an AI scientist haha. But why is it that we can't have a normal conversation about this topic? Why is it that you seem so agitated and emotionally involved in this and seem to attack me personally and focus on all kinds of unimportant things instead of just having an interesting conversation about the topic at hand? Always seems very weird to me
@@WieseTechnology Well, if you cannot believe the evidence of you own eyes and now ears, there is nothing I can do to convince you. But I might point to you towards the work of Marcus Hutter and AIXI working worth Kolmogorov complexity. Compression/Prediction/intelligence are basically synonymous.
@@marcusrobbins854 Oh you can do many, many things. For example actually talking with me and explaining the underlying concepts you are refering to and specifically how that relates to GPT4-o instead of just basically saying "dyor". I could say exactly the same to you but I'd rather have a conversation and share knowledge and viewpoints. Everything else seems a bit pointless
At what point would you say "yes, this is AGI"?
To find out more about the ASI Alliance and why we (as in humanity) needs them, check out the video over here
czcams.com/video/2kwfJosYJdE/video.html
TLDW : No.
Clickbait Title ? Yes.
💚😌
It isn't evidence that the technology is plateauing. Its basically a minor update on the current model. Lets wait for gpt5 before we go throwing big words around, shall we?
We've seen the plateauing cross different LLM's. Also these are not my words but the words of different scientific articles and papers I've made videos about in the past. Also, how about we just talk and discuss these things properly instead of trying to tell each other what we can or can not say, shall we? ;)
it makes decent code if i copy paste in the documentation and get it to read it before doing anything
Are you using it a lot for coding?
@@WieseTechnology im mostly using it for finance python code. id say you need to point it in the right direction and a lot of the time paste in the right documentation but after that it makes things pretty simple also id say gpt's best feature is that it has very shallow depth in a massive amount of subjects so ideally string unconnected concepts together and then combining concepts in new ways is where it really shines. like for me combining exponential HHPS request time outs to algo trading to avoid when the stock market has unfavorable positions to avoid losses would be one example. ie if i make a trade and it was a loss ignore any incoming buy signals for 5 days if i make 2 losing trades ignore incoming buy signals for 15 days stuff like that where they mimic other ideas found in HTTP requests.
i know this technology is being said as being some sort of leap forward. What I have noticed in playing around with this, is that this "latency" thing isn't really a thing. One way a human delays answering a question and provide themselves with a few seconds to formulate an answer is by repeating the question. ChatGPT does this. Try it yourself. All that time responding to you and reading back what you asked is simply giving it more time to formulate the answer to the real question. The tech is cool, but this focus on the latency .. . well, not sure it's all that it's being hyped up for.
Yeah I agree, you can also notice how GPT uses fillers to have more time to come up with an actual answer. In my opinion latency is also not a big deal. Who really cares if the AI answers in 1 second or in 0.3 seconds. Not really anything else but a marketing gimmick
Can we get this free gpt-4.o yet or is it still in beta???
It will probably roll out over the next weeks for everyone, starting with America. But not all functions will roll out. For now, I think only the improved text capabilities will be available with hopefully the voice/video capabilities coming soon after
AGI (Artifical GENERAL Intelligence|) is not to be mistaken with ASI (Artifical Super Intelligence|). You fail to understand this tool is beating top expert level humans in any test we can come up with. Consider theory of mind or other causal reasoning, you think average or median humans even consider deep thought that is beyond intuition. Average performance has been achieved a while ago, they are just patching up the hole with great performance in other modals.
Yes obviously AGI isn't to be mistaken with ASI. Still there are many different definitions for AGI and even more definitions on what "intelligence" actually means. LLMs have great knowledge because they are trained on vast amounts of data. Yet if you dig deeper in certain areas they fail to understand things or reason properly beyond a surface level understanding. While it might seem they are "intelligent" mostly they are "knowledgeable" which are two very different things. For example, no one who wants to be taken serious would claim we achieved AGI because an AI could beat the best human in Go. Aside from that I also haven't heard from ANY noteworthy AI researcher that AGI has been achieved. I really appreciate your input and would love to talk further but I hope you, just as me, understand that you probably don't understand much when it comes to this level of AI research and leave polemic claims and assumptions like "You fail to understand" out of any further conversation :)
@@WieseTechnology My point was that average humans don't have beyond surface level understanding on most things. You are again comparing to things expert level humans have. Which is not the point to AGI, but ASI. And ironically, you mentioned that we have only surface level knowledge in the level of AI research.
@@Quaintcy My point was that neither you or me understand AI/AGI on a level that Yann LeCun, Ilya Sutskever, Ben Goertzel, or Sam Altmann understand it. And neither one of them or other AI experts of the same kaliber have claimed that we have achieved AGI yet. So it's interesting that you apparently claim to understand it better than them when saying that we have (sorry if this comes across as a bit polemic from my side but thats how I understand your statement). And it also seems to me that you have a very different definition of AGI than I have or what most people would think AGI means. We have an intelligent computer system that is very knowledgeable, easily more knowledgeable than most humans in most fields. But that has not much to do with actual AGI. By that definition, you could say a very big encyclopedia with vast amounts of knowledge would be an AGI. These AI systems don't have a sense of self, no intrinsic goals, no sophisticated theory of mind, not an actual world model or understanding of it, and lack many other things that one would think an intelligent being should posses. ASI is also no knowledge of what expert level humans have but FAT beyond. Let's try to use IQ to understand ASI here, even tho of course it's lacking many other fundamental things that would need to be in place. But just as an example to understand ASI. An average human has an IQ of 100. An AGI should then also have an IQ of roughly 100 and an ASI system would have an IQ of 500, 1000, 5000, and higher. For an actual developed ASI we humans would in comparison be basically ants.
@@Quaintcy By your definition, if I just made something that recites the wikipedia article on any topic I ask about, it would be smarter than most people because most people don't know that much about most things. It has nothing to do with knowledge, or the ability to answer questions in a test. It is about cognitive ability, very different thing. AGI requires the ability to use deductive, inductive and abductive reasoning. It requires a world model. Machine learning is about induction, and then manually patched with deduction when that fails. Without real abduction, it is impossible for AI to really learn or understand anything, or navigate the real world. No matter how much data you train it on, you cannot send a robot to wander about in a city and run a few errands, because anything that happens that doesn't closely resemble something in its training data or programming will utterly confuse it. That's the difference. There is no broad set of cognitive tasks that an AI can perform as well or better than humans, even if it can recite memorised answers to test questions.
@@WieseTechnology Right, to be clear, my definition of AGI revolves around the term General. You can either define it median, mode or 75 percentile performance in all the test of intelligence we can come up with and ones we haven't, that include metrics for sense of self, theory of mind, world model. As we define intelligence by those measures, it is the only definition of intelligence we know of, it is intelligence as humans can tell. Surpassing 75 percentile would be AGI, but we have not have developed those metrics. AGI is not official.
Average IQ is only 100 for a reason. IQ tests spatial reasoning, which can be argued is part of world model. We've not set the bar very high for AI. I did find a article claim gpt4 score 85 on the mensa test and claud 3 got a 101. No result on gpt 4o yet. 4o is not behind if claude 3 can do it.
I don't even know what sense of self or theory of mind would test. I do not believe there is consciousness or free will. Which is another whole discussion. Those metric therefore is not included in my definition.
I'm not even going to respond to the appeal to authority fallacy. More interesting if you quote them. I suspect their reluctance is more likely due to polics of such statement.
Then the other half of expert can't even define AGI, or uses metaphysics. Then they rage about the quailia is not right, as if it's something that can be nailed down, or is even the same between people. They argue based on a feeling.
yeap
🙌
❤
🙌
This is the dumbest thing I have heard today. Even AGI couldn't convince you that you are completely wrong.
So you think that GPT4-o is AGI? Interesting, maybe you care to elaborate on your point and thoughts?
Ideally without insults or polemic language and instead with actual arguments and well founded knowledge. Should be especially Interesting since not even anyone from OpenAI or any reputable AI scientist would claim that GTP4-o would constitute as AGI aside from one guy on X. But very interested in your point of view. Having an actual conversation is usually something I prefer :)
@@WieseTechnology Actually if you are so smart, dont film your edits with a mobile phone ^_^ Maybe fix the videos sound and maybe try to talk a little less histeric.... i had to silence you for a splitsecond because you were fucking screaming, compared to every other youtube video i watched before
Und wenn das alles hinbekommen hast, dann kannste noch an deinem reudigen englisch übenAsk chat gpt how to do these things.....
@@darthfather1294 Look this is exactly what I mean. I never claimed I'm smart and I'm definitely far from an AI scientist haha. But why is it that we can't have a normal conversation about this topic? Why is it that you seem so agitated and emotionally involved in this and seem to attack me personally and focus on all kinds of unimportant things instead of just having an interesting conversation about the topic at hand? Always seems very weird to me
@@WieseTechnology Well, if you cannot believe the evidence of you own eyes and now ears, there is nothing I can do to convince you. But I might point to you towards the work of Marcus Hutter and AIXI working worth Kolmogorov complexity. Compression/Prediction/intelligence are basically synonymous.
@@marcusrobbins854 Oh you can do many, many things. For example actually talking with me and explaining the underlying concepts you are refering to and specifically how that relates to GPT4-o instead of just basically saying "dyor". I could say exactly the same to you but I'd rather have a conversation and share knowledge and viewpoints. Everything else seems a bit pointless