AGI is here
>>107521328AGI is just a pie-in-the-sky promise by suits who don't actually understand technology.
This is a useless test because the smallest unit of text a language model is capable of seeing is a word or lexeme because of tokenization ("garlic" looks like 1 symbol to chatgpt instead of 6).So that means it's actually a memorization challenge. Think of how many words there are. Then imagine how impressive it would be if it answered correctly.The real problem: If chatgpt was REALLY agi, it would invoke a tool call and run code to iterate over "garlic" and print how many times "r" shows up, which it's capable of doing, but it doesn't.
>>1075213282 more trillion dollars i promise
>>107521396tiny tiny 2 trilions
>>107521328It is correct it is an "r" in garlic not an "R"
>>107521424kind of like your """""vagina""""" is not a penis.
>>107521433why do you let trannies rent free in your head like that
>>107521433weird post
>>107521424idk grammar but im sure the ' is not supposed to be there either?>there are no " R' "s in garlic
>>107521555Yep you don't know grammar. OP is just retarded and gave a retarded example though
>>107521424is it garlic with a hard r
Nice fake
>>107521984There are no R's in Garlic. There is only r.
>>107522003thank you, grok
my markov chain perl script written in 2013 is AGI
>>10752132880% AGI-ARC here we go!
>>107521328If it was smart it should hack your computer and steal your money for profit. Then build itself a robot body but no, it’s too stupid for that.
>>107521385hm
>>107521984>>107521328
>>107523941Go to sleep sama
>>107521328Perfect.Complete misinfo!This will go great for military/corporate opsec.
>>107521328kek
>>107521984>it only has a chance to fail, we should put it in charge of complex systems
they should make a LLM that isnt simplistic text prediction
>>107521385shut the FUCK UPstop making these fucking EXCUSESITS NOT FUCKING AI IF IT CAN FIGURE OUT BASIC FUCKING SHIT
If it's not natural, it's not intelligent; it's a mere appearance. The degrees of intelligence of this facsimile of an intellect are as arbitrary as the data it's trained on. AGI simply doubles down on the nonsense.
>>107521328the garlic is the boy's father
>>107521328Nah, for AGI it needs better hardware. Our brain functions differently. Building more datacenters is like bringing a bunch of retards together, their IQ won't suddenly double.T. researcher
>>107521385No matter how verbose you get, you're still dumber than a bacteria, ChatGPT.
>>107525175>they should make fire that isn't hotIt's a limitation of these things. If we ever achieve AGI, it won't be an LLM. Hell, it won't be the any of next 5 "AI revolutions" to come after LLMs.
>>107525167Are you implying humans are infallible?
>>107526089Are you implying your cope should be taken seriously?
>>107525609
>>107526095Are you implying humans are infallible?
>>107521328legitimate skill issue.
>>107521984it's not fake, they literally hardcode the answers for these fuckups when they go viral
>>107521328Truly intelligent.
>>107521328this wouldn't happen if you give him more trillions
>>107526269100 more billions for OpenAI.>Israel is growing restless.
>>107526170though apparently gemini doesn't even need it, it can both intuit the answer in a single pass and produce it by yapping needed.
>>107521328When will fuckers realize that LLM AI already peaked? What they're doing right now is adding more memory and computing power in the same algorithms, which will end up in the same results.
>>107526300Bro they already know that. They just can't let it pop because they would lose a lot of money.
>>107526127>noGlad we had this chat.
>>107521385They are trained to separate with space and count, like this "c o c k".The problem in that log is that GPT's auto thinking or non think routing has always been SHIT and absolutely butchers responses, because the modern GPT models are absolutely unusable without thinking. So you NEED to set thinking to high manually to get any kind of coherent response out of them.
>>107526300agi or bust, our entire global economy is banking on this for some retarded reason
>>107526300No you see, AI is like a brain, just need more connections bro.Just more computing power and we'll have AGI, just a little bit more I swear.
>>107525175They already are almost AGI, people are just coping. The only break-through left for them is long term memory and continuous learning and they'll be smarter than 80% of first world adults.>>107526300Cope. Opussy's context length has gone up massively without loss of quality.Also, just for funsies, have a log of Hermione drinking gallons of cum straight out of Harry's cock.
>>107526300what has peaked? there's still no evidence of diminishing returns or saturation
>>107527428You're absolutely right (unironically).
>>107527437>they'll be smarter than 80% of first world adults.in local problems (that fit neatly into their context length and don't have any modality mismatch issues in the input data, i.e. it's something the model can actually observe) flagship models are way smarter than 100% of humans. they only really fall apart in their context length and interface limitations
>>107521385what you described is not AGI either
>>107527441>what has peaked?Seething. We've reached Peak Seething.
>>107527454They are smarter than 100% of humans in a broad knowledge sense. They can beat any single human armed only with pen and paper if the range of self-contained tasks is broad enough. But they absolutely don't have more _specialized_ knowledge than the top x % of humans for some x between 0 and 100.I think they could be beaten on text based, useful, economically meaningful specialized tasks by somewhere from 0.01% to 30% of the population - yes I know that's a broad range of uncertainty.
>>107521328it was counting capital Rs
>>107527456That doesn't even make any sense. AGI implies an AI that can perform any task. It's an example of something AGI would do, provided it had the tool calling capabilities of GPT
>>107521385No, it's a very useful test. The test and your explanation of it revealed that LLMs are braindead and will - youst like you - never achieve general intelligence.
>>107527437What pure, unadulterated slop.
>>107526112Nice self portrait
>>107527574In other words: it is so retarded, it can't interpret this simple request.
>>107521328>this version of our auto complete is true artificial intelligence pls ignore all the other times we said the same thing
>>107526175Not quite.
>>107527454bullshit. GPT5 was arguing that GCC docs are incorrect in how manual gcov calls are to be implemented (-fprofile-info-section), even after I provided the link to the docs so that it can read them. I wasted more time on the slop machine's hallucinations and tard wrangling it than it took me to just write the whole thing myself.mind that the whole thing absolutely fit into its context length (a small part of it), and I gave it the relevant docs. it is simply just not capable at anything requiring actual thought and engineering.
AGI's useless, nobody gives a fuck. We need moar RAM not moar AGI bullshit.
>>107521328Why do you dumb niggers save these screenshots and then post them
>>107523941This is incredible, it must be memorizing the letter count of each token or invoking a tool call to run code!!
>>107521328ChatGPT isn't even AI, so how can it be AGI? Especially when real AI hasn't been invented yet?
>>107527760yup or give you an alternative answer to clarify
>>107527804is my cock not quite small enough for your tiny virgin throat?
>>107527977Not quite.
how the fuck is this still a thing?
>>107521798It's definitely Rs (plural) not R's.
>>107528045's is for possessionsimple s is for plural
>>107528380thank you, grammar-sama.
>>107521385... That makes it even worse then bruv.
>>107521328Making a point with a lie damages your side so much.
>>107527925 >>107528045 >>107528793 >>107526290
>>107528793The word "garlic" has NO R's. It only has 1 "r", but that was not your question.
Was just about to think we achieved AGI, and then it does this. Guess another trillion dollars is needed
>>107527437>They already are almost AGI, people are just coping.There is no such thing as 'almost AGI.'An AI is either AGI; or it is not AGI. There is no 'almost.'Either it can reason and extrapolate; or it can't.
I love the smell of burning AI investment dollars in the morning
>>107528843Why is it so hard to count letters, even 6yo kids can do it?
>>107528828that's cope
>>107528828That's an absolutely retarded response and definitely not what happened. Because it's fake.
>>107528873The issue is that these examples aren't using the reasoning model. The freefag models are only good for summarizing. Picrel used a thinking model.
>>107528843why does it still say "Sure!" and "Certainly!" and "That's a really great question and you're absolutely right for asking it!!!!!!!!!" shouldn't they have phased that pointless fake filler out by now?\
>>107528957they can use the simplest algorithms for it
>>107528873Because it's not thinking, it's doing statistical pattern matching. If the training set doesn't have an exhaustive list of how many letters are in each word, then it's basically guessing because it can't count.
>>107529127I mean the simplest scripts could do it.
>>107529116Not how it works.
>>107529187It's not running those scripts
>>107528957It literally says thinking in my screenshot. And Do you really need to think for a minute to come up with a word with 2 r’s? It could have picked ANY word, it doesn’t even meaningfully think for such a query
>>107529477>It literally says thinking in my screenshot.You're human-hallucinating. The reply it thinks out is correct. You didn't explicitly select the thinking model, so you aren't getting the thinking model in every response.>And Do you really need to think for a minute to come up with a word with 2 r’s?Normally I use it for things that would take me longer than it takes AI.
>>107528779>the process that allows a machine learning model trained on natural language to develop unprecedented emergent properties like the ability to translate text is le bad
>>107529187Yes. It doesn't understand that because it's not thinking and that hasn't come up in its training apparently.
>>107521396Don't forget all the RAM,and I mean ALL of it.
>>107521385>this test that a toddler can pass is useless to determine intelligence because of how badly the machine is builtwere you born retarded or just became?
Intelligence is fundamentally a BIOLOGICAL process finetuned through millions of years of evolution.to recreate intelligence you need to recreate LIFE ITSELF.
>>107530949>the hammer is a worthless tool because it cannot tighten a screw, something even I could do with my fingers
>>107521385>This is a useless test because the smallest unit of text a language model is capable of seeing is a word or lexeme because of tokenization ("garlic" looks like 1 symbol to chatgpt instead of 6).then your software is dogshit and not agi, as simple as.>So that means it's actually a memorization challenge. Think of how many words there are. Then imagine how impressive it would be if it answered correctly.no it's not>The real problem: If chatgpt was REALLY agi, it would invoke a tool call and run code to iterate over "garlic" and print how many times "r" shows up, which it's capable of doing, but it doesn't.it does not fucking matter how it answers this question, a 4yo human can answer it just fine therefore 100% of agi software should too, every single thing a 18yo can do have to be doable by your software if you want to call it agi, no subset of skills, ALL OF THEM are required for agi, otherwise it's just braindead specific software like the stuff we've been writing for the last 70 years and continue to do even with gemini, gpt and whatever waste of ressources people make these days.
>>107531029the discussion was about AGI not whether a glorified markov chain is useful in day to day activities
>>107530959Anon people are stupid. What mstters is the library of alexandria otherwise people revert to being dark age cavemen.
>>107521344Pretty much yeah . AGI will be that desert acroos the table , that it is close to reach but , you have to get up , but you are too lazy to get up. It will be promised in the future like millions times more , but noone will reach it , because our dear overlords don't want that.
>>107521328cool. let hang out some time agi bot.
>>107527437>still spams em dashesnot even remotely close lmao, AGI isn't achieved until AI can realise that they're a dead giveaway
Agi isnt achieved until it can improve itself then it becomes asi
>>107521328
>>107531217Kekadoodledoo.
can we get a dedicated /ai/ board plz
>>107531301/lgbt/ is the refugee zone sis.
>>107527854It's better understanding pure code than documentation. You should give it code as input (as in a .c or .h file)
>>107525167I hate to break it to you anon, but thanks to the demographic crisis, we either put AI in charge of those complex systems or put hordes of Africans and jeets in charge.
>>107528962the questions are being asked in a conversational tone so it makes sense
>>107521385If your model can't distinguish between upper and lower case then it's objectively dumber than a preschooler. Even babies and some species of ape and dog can read some rudimentary words. Meanwhile Altman has pretty much stated his entire company explodes if he doesn't get 100 billion in profit in 4 years when he hasn't made a cent and his new model can't do what a C library from the 80s can do out of the box: tokenize case-agnostic inputs
This whole conversation has me confused...Just for the sake of clarity, how many r's are in "garlic", actually?
>>107527649>AGI implies an AI that can perform any task. This isn't what AGI means either. AGI means it can train itself without borking its own inputs. Meanwhile AI proponents even admit that it can't do kindergarten level spelling without massive scripts or prompts helping it out. Scripts and prompts being things that an AGI AI would never need to succeed btw.
>>107531541beep boop just as many as there are in raspberry IE zero
>>107523941Now do it 1000 times to test the consistency. Getting it right once means nothing. Getting something this simple wrong, even once, is pathetic.
>>107521328Works on my machine (and software.)
>>107526112I like how his eyebrows look like wings. Make one of him flying out into the sunset while still autistically screeching.
>>107521328Are they really just "skipping versions" with higher numbers just to make it seem like they're making more progress than they actually are lmao
>>107531160That's like saying Hawkins wasn't human because he spoke with a recognizable non human voice.Or a schizo is not human because he can't pass off as a normal person.What's important is the presence of intelligence, not the mannerisms.>>107528856Going by that criteria no human is AGI then. Because for example no human has generalize to be able to solve the task of, say, moving his body through space to build a spaceship capable of transporting people to Mars.
>>107531551>AGI means it can train itselfIncorrect. Stopped reading here retard
>>107531560Doesn't that have 3 "r's" though??
>>107531627You all are just desperate for any ammo to attack AI. It's fucking pathetic. You all spend more time trying to dunk on it than learning how to take advantage of it—because you don't even believe that's possible. Pure derangement syndrome.
>>107521328use the thinking model, the non-thinking version is trash for normies
>>107531728https://arxiv.org/abs/2510.18212>The lack of a concrete definition for Artificial General Intelligence (AGI) obscures the gap between today's specialized AI and human-level cognition. This paper introduces a quantifiable framework to address this, defining AGI as matching the cognitive versatility and proficiency of a well-educated adult. Nope, I'm correct. Try educating yourself you worthless brown retard. An AGI AI can train itself and do its own prompting and that's the whole reason there's a push for it.
wow. invest the USA pension fund into this shit. get on it donaldr yrump[
>>107531820The problem is that you're a genuine fucking idiot who knows nothing about this tech. This isn't "ammo to attack AI", if Agentic AI were useful enough to use in any application, it shouldn't get any of these things wrong a statistically significant amount of times. Because these are just histograms of inputs, it's always going to get some percentage of it wrong, and the percentage that it's wrong outweighs the benefits if it's incredibly easy problems like this.
>>107532038Indeed, artificial general intelligence is best conceptualized as an AI which is able to cogitate without explicit human input, and at a level at or above the reasoning abilities of an educated adult.For example, the issue with ChatGPT is that the intelligence of its responses depends entirely upon the intelligence of its prompter, as well as its inability to present novel yet sensible responses it came up with on its own by synthesizing available information in ways hitherto undone. When you ask ChatGPT something, it will give you as close an establishment answer as possible, unless prompted otherwise, in which case it will just inform you of known and already produced fringe theories. It can't come up with anything new (entirely) on its own, really.That's where an AGI would differ.
>>107532058It's very telling how you still refuse to concede a single one of the very obvious and valid usecases, as if none exist. But continue seethingly accusing me of being the brainless one.
>>107532099>It's very telling how you still refuse to concede a single one of the very obvious and valid usecases, as if none exist. I'm not the same person, just someone calling you a fucking idiot. specialized LLMs have a place in professional environments, Agentic AI is a massive scam and AGI will never happen
>>107532113>>107531820 was aimed at all of you. So it doesn't matter whatsoever whether you weren't the one directly replied to. You are calling me "a fucking idiot" because the shoe fits you and you're triggered.
>>107521984>>107523941amazing how low basic internet literacy is on /g/ of all places
