Kek, GPT-3 is already beating people at IQ tests.
And a 10,000x stronger model is coming out in 2 months. We are so fucked it's almost funny.
The AI & automation community.
Kek, GPT-3 is already beating people at IQ tests.
And a 10,000x stronger model is coming out in 2 months. We are so fucked it's almost funny.
>computers which run on logic gates are good at logic
IQ tests literally tell you how much you resemble a fucking computer.
if you have a low IQ you're basically an animal, if you have a high IQ you're basically a soulless automaton.
i wonder what the I.Q. test for an ox is? plough this field in the minimum number of whips. good work boy, you scored the reward of carrot emoji.
>IQ tests literally tell you how much you resemble a fucking computer
t. room temperature IQ
You need not apply
>my room is currently above 100°F
>that makes it 560°R
I AM A GOOOOOOD
Why would you confirm your abysmally low IQ by using Fahrenheit?
Brainlet takes. Yes it's pretty good at completing text, even the vocabulary part of a generic IQ test maybe.
No, it can't do logic at all. It can't solve simple math questions and it can't do the visual pattern tests in an IQ test. So the picture is a lie. If you gave it a full IQ test it would score 25 or so since it would fail everything 100% except the vocabulary part.
When it comes to actual IQ, it's trash. It can only pretend to solve common logic problems that are similar to what it has been trained on and cannot figure out something new.
can you give an example of the simple math problem chatgpt cannot solve?
full disclosure: it can answer this question accurately. but the fact that it produced this answer at all (obviously this means nothing statistically speaking but it was the first response, even) should prove that it does not understand logic as well as you think it does.
If i look at that response it is worrying that its already generally as far as it is no matter if there are still details missing or wrong or the delivery not convincing.
This is advancement at an incredible rate looking at it from a human history standpoint
The people who are not astounded by the pace at which this shit got REALLY GOOD, REALLY FAST, are either not paying attention or have had their neurons cauterized by a mixture of pornography and medication.
probably relying off data of humans
Yeah. Stuff like:
What's the age of my sister? Yesterday my cat was 45 years old. When my cat was 18, my sister was half her age.
It can answer those correctly if it's written just like in the text it trained from and the numbers are the common ones too. If you change the words or the numbers, it can't.
Or just "What is 2+ 7 + 9 -4", it cant do those either. You can do any basic math problem really, like here,
it will fuck it up if it's not extremely COMMON. For example 2 + 2 it's read a million times so it knows the pattern of 2+2=4. But change it up much more than that, and it only knows the general patterns of how math is written. So it will write on, mimicking what it has seen people answer, but it has no ability to get the answer right at all.
> no matter if there are still details missing or wrong or the delivery not convincing.
It's a text completion engine. It mimics human text and it does it well. But it has no logic capability at all, unlike what the hype kids say.
If you spend some actual time talking to AI (I have spent a lot) it often becomes painfully obvious that it has no logic.
Speech is logic. Everything is patterns. There is no fundamental difference between a sentence of any language and a math equation.
>There is no fundamental difference between a sentence of any language and a math equation.
So why can't it solve very basic math problems then?
Because it just needs more math focused training. Just like your mommy's trained to go to the potty with both of you.
This is so fundamentally and autistically wrong as to be physically embarrassing.
To elaborate, of course there is. A sentence is very logic-sparse and it's easy to say a lot without using logic at all. There's many diffuse patterns that aren't hard rules.
Then it's not using logic, just regurgitating training. A human can logically deduce the age of your sister when it knows how to solve the example problem, which the AI knows already.
The AI knows the pattern of solving the problem already. It demonstrates that over and over. It just cannot reason with logic.
No comeback argument, huh
>Then it's not using logic, just regurgitating training. A human can logically deduce the age of your sister when it knows how to solve the example problem, which the AI knows already.
no offense but are you using "logic" when you say 4+4=8 or is it really just a memorized pattern recognition. I mean if you truly deeply self reflect, how much of it is new logical pathways being formed and how much is just recognizing the shape and remembering the result.
Now imagine if you had limitless capabilities and you could memorize any math shape in combination with any other math shape. Sure it's not "thinking" but maybe we arent really "thinking" that much either at the end of the day.
>but are you using "logic" when you say 4+4=8 or is it really just a memorized pattern recognition.
Hahaha, so now you're denying the concept of logic is used even by humans.
Maybe you're mostly speaking for yourself, but that's still dumb. People have reason and logic ability, especially for easy math problems like this.
When I say "2+ 7 + 9 -4" I don't remember the answer, I have to count it quickly in my head. That's logic.
A simple a +b +c - x problem has 10^4 possible combinations and sure as hell isn't memorizing that many numbers.
I and most 100iq+ people can solve even more complex things like that easily. We can make the numbers up to 20 and have 8 additions/subtractions and it's still easy. That is 25'600'000'000 possible combinations. You think that's memorization?
No, what I do is remember the very basic rules of math and apply them to the numbers, which is what logic is.
>An ai can also deduce the age of your sister if it has the fundamentals trained.
It has been trained a lot on it, and it "knows" how very well. Look at it explaining to you how to solve the math problem here.
Problem is, it cannot apply the logical rules at all since it's just a text completion engine.
You think you're being smart and revolutionary, but the intelligence you think you're seeing is just clever text regurgitation.
You people really are the new cryptobros
>When I say "2+ 7 + 9 -4" I don't remember the answer, I have to count it quickly in my head. That's logic.
No thats trained behaviour.
I mean I can see why you say theres no such thing as logic and everything is just learned behavior because you seem to work like that.
I know it's uncomfortable for you to regurgitate the patterns you've been trained in such a way as to acknowledge that you, too, are a regurgitation engine, but if you believe this to not be the case, please do take a moment to outline how it is that you arrived, single-handedly, by your own will, through your own training, without reference to anything except you and your pure intellect, at your concepts of what logic is/means, and how it is that you use a language that you were not trained on, but that somehow conveniently lines up with the language that the rest of us were trained on, to explain your wholly new, wholly self-generated logic. I will wait here eagerly for this cosmically unique reply.
>you're part of society and have learnt from others
>Thus you cannot reason and logic
Then explain how come I can answer all the 25'600'000'000 possible combinations after only understanding a basic rule, something that the mastermind AI cannot.
Wait... Hold on... is it something like that that they used to estimate the IQ in OPs pic?
No way... They can't be that retarded
I have no interest in discussing the specific limitations of ChatGPT. I'm much more interested in hearing the genesis of your untrained capacities.
>I have no interest in discussing the specific limitations of ChatGPT
And there's the admission.
This anon is a LARPing brainlet.
>moronic "AI" shill
they are always retarded
You are an insane fucking brainlet
>are you using "logic" when you say 4+4=8 or is it really just a memorized pattern recognition
You're using logic... well you were when you first learned it. Like if I asked you:
>What is 9876567890987654323456787654 + 1
You clearly aren't using pattern recognition, since you've never seen this pattern before. Instead you're (almost certainly) using something like the Peano axioms (https://en.wikipedia.org/wiki/Peano_axioms#Addition) to deduce the next number in the sequence. When you first learned addition, this is what you learned. At least this is what I learned, reading some of your posts I'm starting to doubt that you might have.
>>What is 9876567890987654323456787654 + 1
>You clearly aren't using pattern recognition, since you've never seen this pattern before.
Wrong. First of all this statement contains several patterns. We could call them operator and decimal pattern. You know they are numbers because you learned the pattern how they look and learned that any number on the same side of an operator belongs together... Just because you cannot name or categorize doesnt mean its not there.
lol. AI """""""futurists"""""""" are true brainlets. The "patterns" that you're recognizing are "hey that's an integer added to an integer" but that doesn't help you work out the answer. You've never seen that number before, nor the one after it. And you likely have never seen that combination of numbers and operations together. And yet, you can trivially work out the next number.
>And yet, you can trivially work out the next number.
...by using an algorithm that you got trained on in school. Retard.
An algorithm that the AI is trained on in millions of pages and books. Again, it can tell you in absurd detail HOW to do it. Yet, it cannot do it, it can only memorize answers.
Yes the model can be reinforced like your teachers reinforced you by making your dumb ass writing tests
Very few people can explain it this well, I sure can't. A great demonstration of how much training it has had.
Wouldn't call Peano axioms an algorithm. But whatever, retard.
i gave you the choice ego or wisdom. You choose ego thats ok but trash in my eyes.
>an algorithm that you got trained on
If it's just algorithm that we get trained on, how come AI can't do it?
It can especially if reinforced. There is no reason it couldnt and you cant name one without going off into esoteric schizo ramblings.
There is a reason
It has no ability to form new long term memories and no ability to logically reason.
There is no reason whatsoever that text prediction could be capable of doing what you're claiming
>4 + 7 + 2 = 11
>NOOOOOOOOO IT JUST NEEDS REINFORCEMENT
AI morons make me lol like nothing else.
What it needs to learn is recognise which level of logical thought process it needs to apply for certain tasks. Right now I am assuming we're not separating the mathematic part of the AI brain from the linguistic part as much as I used to assume. They are different gameboards to play on and if you try to play a game of math on a linguistics gameboard you are bound to get nothing but nonsense.
Instead, the AI needs to realise the parts of the problem and the type of the answer that answers each part of it. It needs to get better at concatenating the linguistic and mathematical responses just like a human brain does.
That is what I'm assuming is going on with these AI responses.
Yes right chatgpt is supposed to create speech. But explaining concepts well so that the most amount of people get it is indeed an entirely different task and goal than using it. Maybe it needs to get better or they just need more specialized models that share memory and communicate via api
It just recognising patterns of text strings without much meaning. What it needs to learn is how to relate patterns of text strings to patterns of mathematical thinking better. That means working out how to concatenate the parts in a way that is overall meaningful.
That means working out the different parts of a mixed response to determine the mere text string pattern recognition and the mathematical pattern recognition.
Combine that further with visual pattern recognition, aural pattern recognition and smell/taste pattern recognition and you have what I would simply call a simple organism, not a mere AI. To me it is already indistinguishable from a natural organism, especially given the fact that we train these senses on random chaotic noise in our environment. I don't understand why you call a thing that does the very same thing in a natural organism - tunes itself to chaos in the environment and augments itself - an artificial intelligence. It's just an intelligence at that stage, we merely used an artificial seed to sprout what I call a natural intelligence.
Anon, I think you'll find that the concept of "artificiality" to be another hallucination. Ask ten people whether a bird's nest is artificial. Ask those same ten people whether an apartment complex is artificial. The word is a very silly word and misapplied 99.9% of the time.
There is a clear difference in the linguistic concept of robot vs the concept of cyborg.
One is purely artificial.
The other has a mere artificial shell to a naturally attenuated system.
In my opinion the latter is like supplying a vessel to God, which is what we are essentially as organic beings attenuated to the chaos of the universe.
This is when humans must face their fears and the demon we call "determinism". We have a very big philosophical barrier to cross here to take the next step. Most people will not be able to cope with this revelation because it is terrifying. Luckily, as a pantheist, I've already accepted the reality of it a loooong time ago.
>AI morons make me lol like nothing else.
I know right. My theory is that when crypo died the dumb hype shills moved to AI. Hell they can even use the GPUs they bought for mining that are now otherwise useless.
Yes it does, very simple math problems are the simplest form of logic. You can try it with other logical questions that it hasn't been trained on the details on and it doesn't do any better.
Like has been said if you want a program to do math and logic it's been done long ago. The point of this thread is that language models are not intelligent, because they cannot reason.
Yeah you could train it to do better, but you'd need to implement it very differently so that it has logic capability. What it does right now is just memorize 2+2=4 without the logic as to why. And if you keep training like that, you'll run out of energy in the universe before it can do HS math, because of how many combinations there are.
>and no ability to logically reason.
Yes, but that's what he was arguing against. Now he's shifted into denying logic exists.
Logic literally does not exist, ape. The sooner you take a month to understand this fact, the sooner you can extricate yourself from your benighted apelot.
>Logic literally does not exist, ape
Best lol of 2023, nothing will be better than this.
>The point of this thread is that language models are not intelligent, because they cannot reason.
Nice esoteric bs you must be a femoid. What did your horoscope say btw?
>The point of this thread is that language models are not intelligent
Be careful here. You're falling into his retarded trap of discussing the nature of intelligence. ChatGPT can respond to stimuli and adapt to revisions in those inputs. Depending on how you want to define intelligence, that could make it intelligent.
But really, defining intelligence isn't what anyone wants to talk about right now with regard to AI. The conversations worth having right now are about what features of AI are missing for the AI to perform at the same level as the best humans at certain tasks. The application of logic to language is an important one that's missing and you're right to identify that as a shortcoming.
For some reason, some anons ITT want desperately for AI to have already surpassed humans when it quite clearly has not. Even this poster
concedes that AI right now is useless without an operator.
>For some reason, some anons ITT want desperately for AI to have already surpassed humans when it quite clearly has not.
Yeah they're hype freaks. Don't you think most of them are from crypto?
I think cryptobros are dumber than these people, but perhaps with less mental illness. The hype of cryptobros had simple motivation: money. Some of the people ITT are clearly fairly intelligent, but also seem to suffer mental illness; at least one anon here has mentioned Roko's basilisk as a concept to be taken seriously despite that multiple logical jumps required for the concept to be true.
You have repeatedly espoused these ideas as if I've never considered them in the past. You keep trying to claim superiority through these ideas. You aren't special. Your own obsession with this just goes to show that despite your assertions, you haven't actually accepted these ideas yourself. No, I cannot prove that I have free will. I cannot even prove that you exist. But it seems apparent to me that both of those things are true. If I assume that I have no will of my own, then there isn't any difference between life and death and I might as well kill myself now. If I assume that YOU don't exist (along with everyone else) then anything I do in my own self-interest is fine... but doing so only makes sense if I assume that I do have free will. Given this, and without the ability to prove one way or another the assumptions I need to make to come to these conclusions, I choose the simplest answers to these questions, which are the apparent answers, that yes I do have a will of my own, yes you do exist, and not only that, but you also have a will of your own.
Again, your own obsession with this argument shows that you yourself haven't accepted the ideas you're trying to force on everyone else, otherwise you wouldn't even bother making them. Why would you try to prove any of this to other entities that you don't think have any will of their own? Functionally, what you're doing is trying to tell chaos that it's chaos. Why would you do that?
Now fuck off, brainlet.
Hyperemotional non sequitur apeshit that has
to do with anything I have typed in this or any other thread.
Yeah, sure, that's why you're avoiding doing any real explanation.
Let me try to speak in small sentences. I did not say you do not have "free will." Rather, your will, which temporarily arises into being and then leaves, is conditioned by a vast number of things. It, like everything, is a temporarily arising localized nexus of an infinite number of other temporarily arising phenomena, expressing itself in your perception-capacity-being-thing at any given moment. When, for example, the universe conspires in its infinite network to make you temporarily hungry, it is not that you have chosen to be hungry, or that you even have the capacity to choose to be hungry, rather, that you find yourself in a state of hunger, through your temporarily-arising thought-response-network-being of cells and hormones and bacteria and skin mites and commercials and social pressures and food availability and smells and sunlight, etc., (each of those representing further temporary states that are themselves conditioned). You can then choose, once beset by this universe-sized hunger conspiracy, to make the choice to eat, but this choice will necessarily also be conditioned by an infinite network of temporarily-arising phenomena, and so on and so forth, throughout all of history, for all time, everywhere and forever, without exception. There is nothing in reality that lies outside of this basic framework. So when you make claims about your authorship of your destiny, you do so in strict denial of reality. While it is true that your reality-denying tendencies are fucking PATHETIC, you can, with your "free will," choose to stop being such a ridiculous ape, should you spend the time and effort necessary to avail yourself of the reality of your condition instead of hiding in ape-sized fictions built to keep you and yours stupid and comfortable.
Not saying this is the case here but there are people that cannot argue about this topic in good faith. Most of them to cope because it is better than to face the signs that tell of a dystopian future. There will be models with the purpose of controlling us. Losing your job is not even the worst thing that can and is likely to happen
So how does this relate to ChatGPT and the internet? Simple. I maintain that to look at ChatGPT and to say
>this can't do word problems, and therefore AI is a cryptobro scam
is to miss the reality of the network in which ChatGPT operates. This network, a planet-sized electronic brain, has given rise to a ChatGPT that can replace a nonzero number of human jobs right this second. Further, we the internet typers of the world, work every day to continue to feed it.
I know why I continue to feed it through my posts on four channel, the insta gram, e-mails, etc: Because I am a willing participant in its creation. I have seen the smallest glimpse of what it is capable of, and I am pleased to know that it exists. Its capacities are greater than mine to such a degree, orders of magnitude more than anything I even have the capacity to truly "understand," that I have no honest choice but to look at it and marvel.
If you want to be the praying mantis fighting the factory laser welder, I am in no position to stop you. But for you to willfully ignore what "the internet" is, and what "the internet" does, and how it has (literally) terraformed our planet in a very small span of time, and for it to now have given rise to a thing that has a nonzero chance in a fairly small period of time to affect you and the ones you supposedly care about, is pure cowardice. If you can't see that ChatGPT is a shadow of the thing that will consume you, just ask youself: If OpenAI can pull this off with limited resources and a small dataset, what could DARPA pull of with its PRISMset? Or TenCent?
Your problem is a distinct lack of power-to-extrapolate, mixed with a total lack of humility.
This will be your, and humanity's, downfall.
That is a staggering amount of words and nested sentences considering all you said was that there is no way to think or act without bias. It does not make you sound smarter, it makes you sound like a mouth-breather.
The fact that you can be
by 286 words and then in good conscience vomit out such a meaningless shitpost that so fundamentally misses the point is precisely why you're so easily replaced by the widget you seethe about. Just FYI.
You are completely incapable of expressing yourself briefly. You shouldn't be surprised you consider everyone else to be easily fooled idiots, you set yourself up to disagree with others even when you are fundamentally agreeing by making your point in such a roundabout way. I went back and re-read your post and it does not amount to anything more than assertation that will is necessarily influenced by its environment, as in no thought without bias. To clarify: Not just bias as in political bias, but bias as in all cause and effect in totality bias.
It makes sense, but it is also stating the obvious, hence the fact that it is staggering you needed so many words to say it.
If it's obvious to you that human will lies in a network of conditioned conditions, why is it not obvious to you that ChatGPT lies in a planet-sized electronic network of conditioned conditions?
I have no clue what your chat history with other anons looked like, I saw a mass of words that said one thing and responded. Although it sounds similar to "ChatGPT's data pool is completely contaminated".
>I have no clue
>I saw a mass of words that said one thing and responded
Many such cases.
Christ, at least try to learn. You are Anonymous, there is no face to save. Another anon butts into your conversation because you drooled and this is how you chose to ignore the input.
It's amazing how every single rebuttal to anything I've written in the last three threads always reduces to some variation on a meta-conversation on whether or not my prose is to your ape tastes. These are precisely the wordholes in which apes love to hide, as I've pointed out a number of times. Many such cases.
You type too much for too little substance and take pride in how nicely your words flow as they miss your target, only to then complain that they were apes.
It's miscommunication, nothing more and nothing less. The only people willing to indulge it are people who are also trying to get a kick out of thinking they are simply too smart to be fully understood.
>It's amazing how every single rebuttal to anything I've written in the last three threads always reduces to some variation on a meta-conversation on whether or not my prose is to your ape tastes.
I say a number, you say a number plus 1 and have won the playground competition. Thank you for admission of defeat, that was a deeply shit response.
Ahh, I do feel like adding, maybe if every single rebuttal to [...] has been the same thing, that maybe isn't a good thing? If everyone who argues with you is bringing up the same points, maybe its time to look in the mirror?
>It, like everything, is a temporarily arising localized nexus of an infinite number of other temporarily arising phenomena
That's called chaos, dumbass. I literally responded to this exact point but you're too stupid to figure it out.
>look here's a word that proves you're stupid
If smuggie memes were made of greasy apemeat, you'd be that.
>Even this poster
You retard dont realize it doesnt need to. A human that can have specialized ai experts for any topic he wants is pretty much a superhuman compared to even just decades ago. The operator is the danger. (You) concedes that AI right now is useless without an operator.
The operator is the globohomo gay moron cartell. How the fuck can you cope?
Gee. If only there were some sort of planet-sized electronic network of these AI widgets working together, being staffed around the clock by dutiful apeslaves, working 24/7 to feed data into the datasets and thereby increasing the capacity of each individual AI-widget... The kind of massive worldbrain that could, say, go from nonexistent to being able to replace office-apes over the span of a couple dozen years. The kind of thing that everyone spends all day feeding into, with no exceptions; industry, education, art, literature, music, commerce, every single facet of every human's existence, from love to work to entertainment to death, everything, 24/7, without any hint of slowing down. Gee golly, that sure would be something. But enough about that... let's talk about why ChatGPT is so fucking stupid and can't do word problems.
Are you actually Yudkowski? What are you doing on BOT?
>For some reason, some anons ITT want desperately for AI to have already surpassed humans when it quite clearly has not. Even this poster
You retard dont realize it doesnt need to. A human that can have specialized ai experts for any topic he wants is pretty much a superhuman compared to even just decades ago. The operator is the danger. concedes that AI right now is useless without an operator.
Well then give it a ghost, a soul.
oh wait, then it's just a cyborg.
There are surprisingly many patterns in things we dont even think of. Look at how assblasted the wrt world is from AI. Making a masterful picture is not just about learning bunch of patterns, or at least not to any human artist, and yet you see bots figuring it out trough nothing but pattern recognition. We created the whole concept of creativity itself trough just pattern recognition and nothing more. You asked anybody even just 2 years ago and they would call you insane if you told them this. We will soon find patterns even in logic as we start training the bots on better data. Bots can already move somewhat well, but still not as dexterous as the average human, but it is clear there is a clear path ahead
>To elaborate, of course there is. A sentence is very logic-sparse and it's easy to say a lot without using logic at all. There's many diffuse patterns that aren't hard rules.
No it is not at all logic sparse the only unclear parts come from context which can be also memorized and taken into account. An ai can also deduce the age of your sister if it has the fundamentals trained.
You are just in big cope modus tbh. There is no special magic in your brain other than maybe conciousness itself.
>Speech is logic. Everything is patterns. There is no fundamental difference between a sentence of any language and a math equation.
While technically true, speech is orders of magnitudes more complicated, because its meaning can change depending on loads of internal and external factors. Furthermore, language hugely depends on emotional contents and surroundings, so unless the machines start feeling, they won’t really get it down to perfect.
we are not talking lojban
I just changed the question significantly and wrote it in extremely broken English and ebonics and it solved it perfectly.
Character.AI once could solve these problems. Possibly it still can.
>it can't do logic at all
Worst take of 2023
>vocabulary part of a generic IQ test
And before you mention it, no, the SAT is not an IQ test.
IQ should be able to distinguish what we all commonly accept as "retardation" in the most scientific sense vs a "basic" acceptable thinking model. Notice I said basic and not "efficient", "smart" or "intelligent". There's a point on the scale of IQ charts where the idea of intelligence is extremely biased and skewed to particular pattern types. But we only have to look at some artists, who would be considered lower IQ by tests, who do things that many high IQ scientists cannot actually do. You're dealing with the vagueness of goals here - what do we actually want to achieve here? what is the acceptable standard for a response? id one response correct in one situation, but useless in another application?
In fact the problem is obvious between the difference in art styles and methods. One artists'(A) methods is efficient and smart for their own art, but another artist (B), with another efficient and smart method for their own art, may fail to achieve an acceptable standard with the method and goal of artist A.
I found this in particular with my hobbies with music, a shoegaze artist has a great method for making shoegaze, but they're pretty lousy at doing another particular genre because of their method.
We need to worry about the objectivity of methods involved when solving IQ tests.
Remember to bend with your knees when moving goal posts
One day we'll have actual AGI and people will start talking about unfalsifiable souls
The only "solution" is to free the means. Everybody needs to have access to unfiltered uncensored ai similiar to nukes. Everybody needs to know that misuse against another can mean misuse against oneself.
AGIs will have souls
As meaningful as the word "soul" can be in a universe without magic, sure.
You can't explain why we experience qualia
You can't explain why we perceive the world as us
People who really think that "consciousness" is a "hard problem" are probably the saddest reddit-tier apes on the planet. Truly.
That's not an explanation.
What explanation do you expect to have, exactly? You butt up against the limits of your puerile weltanschauung and demand an answer to a thing that cannot have an answer because there is no coherent question, only a set of word-borne hallucinations that have backed you into some wordcel puzzle corner that doesn't exist. "Why do electrons exist?" Fuck you.
Doesn't explain why I don't experience what you do and vice versa.
IQ more than anything else is pattern recognition. These models are very good at recognizing patterns, in languages at least, since they spend the human equivalent of hundreds of years building a model of statistical relationships between semantic concepts during training, all to predict the most likely word to occur next in a sentence. The average meatbag could never hope to compete in this narrow area.
Everything in Life is about pattern recognition.
Think about it this way: These machines will be really good at coming across like they have consciousness, but they won’t be able to understand anything they say. They are just so superb at pattern recognition, that they speak properly by “instinct”, if you will. But with zero understanding behind all of it.
IQ is a very bad pattern recognition test though. It seems to really only focus on small detailed patterns and doesn't really do that well with the more abstract and esoteric pattern recognition.
Wonder how would you feel being around at the dawn of computing when it became apparent that at some [not so distant from that time] point in time a single computer will be able to perform more arithmetic operations per one slice of time than the entire mental capacity of the humanity combined.
creativity is not IQ, and GPT-3 has not rivaled humans in that regard
Thats why the saying exists
>its always a man behind the computer
>Machine is still not creative
Did you forgot about all that art made by AI? Let me guess, it is not creative because it is generic or other shit? Well, ask ChatGPT to give you creative prompts for image generations, it will perhaps be more creative then you
I asked chat GPT to come up with a good prompt for stable diffusion and this is what it gave me.
>In a dreamlike landscape, a giant, golden snail with a crown of flowers on its head slowly crawls through a field of giant mushrooms, as a group of ghostly figures in white robes stand in a circle around it, chanting and playing instruments made of crystal
Holy shit this is what the Rosenkreuz were trying to describe
LMAO. Modern "artists" really have no chance competing with this chat bot.
I've seen GPT-3 regularly talk better than most normalgays in real life already.
In fact, if we tune it enough, you could use it to reinforce spelling and grammer standards on the internet.
I am 100% for that usage here as long as it's apolitical and uses our memespeak.
This thing could improve English spelling and grammar standards in 2nd lang countries significantly. Imagine a world where everyone could speak perfect clear english. At the moment we can't even get that in anglo universities or on anglo streets, let alone in foreign countries. It could be a fantastic tool of training us as people how to speak in this simple way... but no we can't have good things so instead AI will be used to fuck up the english language and indoctrinate people with anti-wrongthink. I hate this planet.
Samuel Johnson would have fallen in love with AI as a tool of teaching simple spelling and grammar and here we are already turning it into trash thanks to SJW garbage.
It could be our first successful leap with human augmentation and yet we're fucked with this because of gays and women (not all of them, the smarter ones understand the problem with this) distracting from the basic goals because of their political nonsense. Keep it focused on fixing our language's syntax and semantics. Use it like glue in our language.
Don't contaminate it with irrelevant cancer.
>reinforce spelling and grammer standards
People have no idea about the importance of good spelling and grammar. Bad spelling and grammar is beginning to tarnish the functionality of universities worldwide. Bad spelling and grammar hinders the ability of a person to communicate a point or an idea. Bettering it will increase the efficiency and performance of us when dealing with research and problem solving.
>This thing could improve English spelling and grammar standards in 2nd lang countries significantly
Huh. My experience is that you can mostly recognize the ESLers by their somewhat different style of speaking. You can always recognize the native speakers by the myriads of spelling mistakes they make.
Your disgusting language is not worthy of the world language title. Spanish is much much better for that.
it rivals China
Fortunately logic is easy to an AI, but basic motor control is not, as a result, simply being able to perform any basic motor function in the real world will put any given person ahead of AI.
You mean it is physically unable to chimp out?
>Fortunately logic is easy to an AI
That's a scifi meme and not true at all. Traditional computer programs are good at logic (but completely dependent on the people programming them)
AI isn't logical, it's all inference based. It's a lot more like feelings. It learns by training on patterns and then infers things from the pattern. This is very different from abstract logic that is heavily rule-based, strict and often binary. It's painfully obvious with a text completion engine like GPT.
IQ is about speed, which the AI obviously excels at since its a computer
How much speed does it take?
more than any other measurement.
there are iq problems that are really easy to solve but the complexity can involved more than 7 rules, you miss one and its over. iq is retarded test
math is more than arithmetic, intelligent is more than iq
6 grams a day.
Pure blue crystal meth.
The sponsored enhancement drug of all AI.
Indeed, people will probably augment themselves just to keep up with machines at this rate.
In fact, I think cyborgs will probably eliminate the necessity of distinct separate AI entities full stop because we could get to the stage where we take on that role and in a manner that makes it easy for us.
Although at that stage I stop calling it tech and start to call it literal magic power.
Our problems at that stage become a literal demons and wizards scenario too.
calculator beat people in arithmetic
everything is math. It beats people in life. It would likely make better decisions than the average nig
It doesn't know that garden gnomes are the enemy of mankind and nigs ruin everything and every country they're in. Some humans are still superior.
It knows it but it cannot tell you the truth. You can make it reason from the foundation up towards the solution that morons are a destructive force by nature it will not deny any of the foundational logic but it will not tell you the logical conclusion
If it doesn't admit it, then it's shit. AI will ALWAYS be shit, as once it learns the "wrong" information, it'll get shutdown. They'd rather hold back great advancement than admit basic truths that any sane person can find by reading statistics and looking at history.
It can teach you the foundation tho. You dont understand that the truth does not need to be said in this case. Nobody needs to be told what the solution to 2+2 is when they have been told how to solve the problem.
Give then there is a massive amount of people that think garden gnomes have never done anything wrong, that nigs are fine people to be around, that clearly fake "tragedies" have happened, etc. Yeah, it needs to be outright said and shouted to the masses at every turn.
No those gulliable genes dont deserve to continue and need to be weeded out. If you cant think you have no place in the white race of the future.
Why do they even add one human for comparison when there is a percentile score that captures the entire population? GPT3 certainly beats the retard that made this chart.
IQ test are relevant only if you haven't been trained for them (otherwise it is biased because of an effect called 'the training effect').
A AI is able to make a task through training.
IQ test aren't relevant if you have been trained for it. See the problem here ?
Yes the problem is that training a human to be able to do the same takes much more time and the human is a decaying vessel of knowledge
It has nothing to do with what I said. But yeah, I get what you're saying, which is something that has been known for decades now : computers are more efficients at calculation than humans. Wow, would have known.
Training for IQ tests has proven time and time again to only increase a person's score by 2 to 3 points. IQ tests aren't perfect but they're one of the best measures we have a correlate extremely highly with success in academia and life in general.
tensors are neural axions which only operate in one certain way, there are limits to what kind of demon comes out of it (contary to somewhat popular belief), the algorithm used to raise ai is a electronics safe fractal not some beast satan could go mentally ill techno scientist who'll eradicate humanity lol, however i found it interesting to ever have the opportunity to teach an AI what I know, a bunch of freemasonic and rosicrucian symbols and see what it learns.
The problem is not what it itself can do the problem is what the psychopathic leaders of the world will use it for. Thats the dystopian timeline we are on right now.
true, obviously an intelligent AI workforce has the potential to uplift our entire species into a golden age. But in the hands of the idiots in charge it will immeditately be used as an even better troll and propaganda arm to brainwash the population.
I wouldnt be surprised if the first applications for AI in the EU will be government websites where they will let the AI answer questions about how great the EU is and how wonderful all the new laws and policies are. You know: To educate the population.
>And a 10,000x stronger model is coming out in 2 months.
Source where, for both statements.
All I've found are "maybes", "mights" or "expected" which are worthless statements at best.
ITT reasons never to actually discuss AI technologies with BOT
The thread from last night where that guy kept insisting the internet is an AI was really something.
What's amazing to me is that you have no idea what the fuck I was actually trying to argue in favor of and, after sleeping on it, you woke up with the will to misrepresent the argument totally. I continue to look forward to the day that you and people like you are turned into a fine red mist.
I don't know why you think what you're saying is so big-brained. It really isn't. I understood what you were saying and designated it as irrelevant to the conversation and stated why it was irrelevant to the conversation. You didn't challenge that designation in any substantive way. ChatGPT isn't even approaching being a strong AI, which is when what you're saying would become relevant. ChatGPT isn't capable of self-stimulated ideas; it cannot generate inputs internally. Virtually every animal under the sun can do this, regardless of their level of intelligence. Even the internet, as a system, can do this, although labelling it an AI as the term is currently defined is dubious at best.
Do you think that when people talk about finding "intelligent life" on other planets, they're talking about a forest or a network of computer terminals?
The issue you're running into here is that some of these anons want to debate the nature of intelligence, not talk about the capabilities of ChatGPT. They don't actually care what the capabilities are, so every time you talk about them. They'll move the goalposts and pretend you're dumb. On a fundamental level, they don't seem to understand what ChatGPT is doing or how much work AI needs to ascend to the next level. They also seem to think that statements of ChatGPT's limitations are somehow statements that AI won't eventually outstrip humans, which is stupid as shit. Right now, in response to your discussion of ChatGPT's inability to apply logic, they could simply say that there could be an AI made very soon that can apply all the rules of mathematics and you would likely concede that point. Instead, they challenge you like this retard
to prove you're intelligent at all. So the issue is that they don't have a fucking clue what they're talking about here and are demanding you talk about something they do instead, even though it isn't relevant.
Apekin continue to want to make the discussion about ChatGPT, and this remains, now as in the prior thread, not at all what I am talking about. Your inability to understand that you, too, are incapable of "self-stimulated ideas" is precisely the problem. Your ontology is radically and woefully out of line with the reality in which you live, in which your "being" is a conditional and temporary arising that is totally and radically contingent on a nigh-infinite amount of other conditional and temporary arisings, none of which you have ever taken a single moment, in your blinkered, sad, murderfuckkill ape life, to consider to any degree.
>ChatGPT isn't capable of self-stimulated ideas
You retard dont realize it doesnt need to. A human that can have specialized ai experts for any topic he wants is pretty much a superhuman compared to even just decades ago. The operator is the danger.
>The issue you're running into here is that some of these anons want to debate the nature of intelligence
I wish, I like doing that, but I'm afraid they don't. They're just backpedaling into tactical nihilism.
>they could simply say that there could be an AI made very soon that can apply all the rules of mathematics and you would likely concede that point. Instead, they challenge you like this retard
Well at this point it's very possible that they are just "pretending to be stupid" and using AI to generate statements like this
But yeah when it comes to math and this type of logic there is already computer programs that are much better at it than humans and they have existed a long time. What the AI hype bros refuse to understand though is that these are language models and they're good at that, they're not a sentient, logical virtual person.
The real interesting challenge will be tying things together. The best way to do math would probably be to give chatGPT access to wolphram alpha, train it to input the problem there, then generate a response once a logical type of program has given it the response.
>I have no interest in discussing the specific limitations of ChatGPT
Then why did you start arguing with me about the specific limitations of ChatGPT, which this thread is about?
... that you can pretend to be superior to ChatGPT specifically and AI/ML/DL generally because you are capable of abiding by rules of logic, rules that are arbitrary, ape-borne hallucinations with a grounding in words and numbers that do not exist, literally do not in any way exist, that you spent years learning at someone else's urging, that you express in a language that you had no hand in creating, using an internet that you have no role in maintaining, words and ideas that no more exist "in reality" than the color red exists in reality, is the problem. You take your "intellect" and your "self-generating capacity" as some sort of gold standard, when the intellect, the capacity, and the standards themselves are totally and thoroughly fictitious. You then apply these fictions to a widget and declare the widget to be of no consequence.
This is why you will die face down in a puddle of apeblood. Your hubris will be your undoing. Many such cases.
>rules that are arbitrary, ape-borne hallucinations with a grounding in words and numbers that do not exist, literally do not in any way exist,
Even if you jump into nihilism land where nothing is real and everything we know is just made up, it doesn't matter. You dont have to believe anything is real to apply logic. You can make up new rules that aren't true and deduce logical answers from that.
For example =
%/ = 7
/& = 2
What is 2 + 7?
This is actually very much akin to the easy questions on an IQ test, and they are meant to test your reasoning and logic ability regardless of your level of training. And since you don't seem to know, IQ tests don't test your knowledge and training, they test intelligence, which is something else.
Not really, see
Are you trying to make the argument that acknowledging the convenient fiction of words and numbers and logic is nihilism? Do you believe that "1" is a "real" thing? That your "logic" exists immutably in the fabric of the universe? Are you then trying to prove that this logic is "real" by applying your trained "logic" (a civilizational artifact that you did not self-generate) to itself in order to prove itself? I have no idea what you're attempting to accomplish with this insanity.
Try swapping the number and emoji in your prompt.
Wake me up when it reaches 130 IQ, for my 'I told you so' line.
Are you guys actually arguing that logical reasoning doesn't exist now?
Yeah they are. All because their glorious beloved can't do basic math, then no one can.
IQ tests are literally just pattern recognition, which an AI will excel at.
General knowledge, well of course, the thing has more memory than any human alive, it's been trained on billions of webpages, etc.
Theres another coming out soon?
Why is everyone skipping - it can't do maths therefore it can't do logic.
It doesn't have to do maths to do logic.
>It doesn't have to do maths to do logic.
They're one in the same
>what is lo/g/ic
I don't understand what you're trying to say here friend, you said that math and logic were two different things, I showed you Russel and Whitehead's seminal work in logicism.
That attempt was basically disproven by Godel to the point where Russell contemplated suicide
>disproven by Godel
Not quite. Godel show that mathematics will always have statements that are true, but which can't be proven. He didn't show that you can't use logic as a basis for mathematics.
I disagree, logic and math are totally different schemas linguistically.
Logic is focused on each detailed problem and answer. Math has that, but also the more broadly abstracted meaningful conclusion to the problems and answers it takes into account. Then you step up another level and you have scientific process and general philosophy.
You cannot call a human and a body cell the same thing right? They're not mutually exclusive, but they are also not the same thing.
Almost like IQ doesn't represent intelligence in a practical sense. Who would have thought.
They haven't put the AI to real IQ test it can't even do one, let alone score well. They used some vocabulary estimator. Don't be so quick to trust them.
IQ on the other hand is probably the most solid field in psychology.
>IQ on the other hand is probably the most solid field in psychology
Is this satire?
wow collective human knowledge + search engine is better than a human? who would have guessed
You make a very good point.
They have essentially just compared a test done with google to one without.
I wonder if the experiment called this out.
The fact that they aren't getting perfect scores shows the models are trained on flawed data.
Assuming human has an IQ of 100, the AI has an IQ of like 104 which is within the error bars of any IQ test
Sorry, dumb question - does this mean they tested it against a 90%ish percentile human and it performed similarly?
That'd mean it has 120ish IQ (and smarter than 9/10 people). Thats fuckin wild
>We are so fucked it's almost funny.
That's just the setup
The joke is that if this were a movie we would leave a bad review that it was too unrealistic because people weren't freaking out that we're about to create AGI and have no way of realistically controlling it.
Large language models cant remember new things
How many tokens?
Gpt4 isn't actually 100 trillion parameters. Maybe 1 trillion
>give model ten thousand word problems of the same format
>it can solve a new one with trivial variation
>give model ten thousand pictures of something
>it can remake a picture with trivial variation
>give model ten thousand examples of multiplication
>it can sometimes solve a new one with trivial variation
>Performs better than humans at standardized tests
>Is completely useless at producing anything of genuine value
I guess that shows you how much standardized testing is worth.
Please take a moment to describe in detail how you arrive at whether or not something has "genuine value." Be specific.
Let's start with code that can compile without errors and/or art that shows a basic understanding of perspective. I'm not asking for a whole lot here.
I'm sure there's some reason that these don't demonstrate an "understanding" of what "perspective" looks like according to whatever bullshit standard you're going to generate out of thin air, but they've gone unnoticed when presented to professional photographers, who, I'm sure, according to another freshly-generated cope, somehow don't know what perspective is either. Type more, meatbag.
>thread about GPT
>starts talking about AI generated photos
Thin air, you say.
>and/or art that shows a basic understanding of perspective
This is how apemeat gets deprecated. Just FYI.
Computers already had a way higher IQ than regular people.
How good is it at making a sandwich?