Thread 107133495

It's almost 2026. How are LLMs still making such simple mistakes? X told me AGI is only a few years away.

>>107133495 (OP)
Threadly reminder that LLMs will not successfully performbasic reasoning in the year 2000 + XYZ.

>>107133495 (OP)
Because the underlying architecture hasnt changed.
In fact they could well get worse because the data sources have changed. In the start of the LLM boom the internet was full of carefully made human content. Now its increasingly filled with AI slop that gets fed back when they do their whole "scraping the entire internet for training content" thing.

>AGI is only a few years away.
Will never happen, because ...
1. requires resources we don't have.
(Microsoft right now is trying to fund nuclear fusion, I.e they might not have enough electricity)
(AI training on the web will become harder with things like cloudflare preventing it)
(Hardware is not getting powerful enough each generation, advancement is slowing down)
2. They have this wierd idea that hamfisting an LLM Will make it AGI soon.
3. Just like with people, having a jack of all trades and a master of non might not be appealing.

>>107133538
>Now its increasingly filled with AI slop that gets fed back when they do their whole "scraping the entire internet for training content" thing.
Theoretically, if you were dealing with a population that wasn't retarded and with content discovery/promotion algorithms that wasn't jewed, you'd have human selection curating the output, i.e. free distillation, so the model would recursively improve.

>>107133495 (OP)
AGI requires two major components:
- dynamic learning and reasoning (ability to discard new information if it's low credibility or quality especially if it conflicts with existing, more substantial data) without needing the entire internet, as a person you don't need to read the entire internet to learn how to speak or write a story
- limitless context

>>107133571
>(AI training on the web will become harder with things like cloudflare preventing it)
It's also training on more and more faulty AI data to begin with, how the fuck can anything improve in that regard?

>>107133596
>It's also training on more and more faulty AI data to begin with, how the fuck can anything improve in that regard?
I disagree if we find like charts online AI is much less trained on random sites rather it's trained on wikipedia, reddit and many other websites like it. However reddit is a garbage way of getting information but for some reason AI companies seem to find gold.

>>107133595
>dynamic learning
Have people completely forgotten that this did happen before.

>>107133702
Tay is not AI just like an LLM isn't AI. You do realize you can finetune any LLM with 4chan and get Tay right?

>>107133495 (OP)
Because it doesn't know what battlefield 6 is? It's not unreasonable for something that doesn't know Battlefield 6 is a new game to think that 2024 is Battlefield 6, since it was a brand new entry that came out after BFV. The fact that it somehow made that jump in logic is kind of interesting since people don't call 2042 "Battlefield 6."

>>107133707
>Tay is not AI just like an LLM isn't AI. You do realize you can finetune any LLM with 4chan and get Tay right?
No the point I am making here is that it will end just the same way.

>>107133730
No your point is retarded because Tay was never an AI, it was an autocomplete. Tay didn't think, it autocompleted based on statistically likely phrases based on training data. So you may reread what I wrote in its entirety and realize you cherrypicked a minute point to make a retarded assertion based more around you trying to flex you 4chan trivia cred rather than being actually intelligent.

If you really want to see how bad AI is, try asking it technical questions about a fighting game.

It's generally very wrong about video game stuff in general. Which is great because it's very easy to prove.

>>107133646
It's going to be more AI slop on those websites in the future as well, so it's still going to get worse for training.

>>107133509
I really like this but I'm not entirely understanding the riddle here. Based on the context given, Bob is not justified in his reply from what I can tell. Unless it's a matter of semantics and saying you can use the same prime value 3 times (7, 7, 7; 3, 3, 3; etc.) as they are technically different numbers, just repeated. But that can't be it because that would defeat the point of the riddle.
3, 11, 13 meets the criteria unless given in an incorrect order, but that is the answer given by the LLM and you say it's incorrect.

Without using arithmetic to figure out the answer, I posit the most likely explanation is that Bob posed the question wrong. What is my retard brain missing here?

>>107133902
It's a dumb riddle because 3 11 13 is correct for Bob's request. The problem is the AI doesn't treat the question like any normal person would which is ignore it or sarcastically ask for the answer. The problem is LLMs are autocompletes that are biased to give you any answer.

>>107133932
Ah, that makes far more sense. Yeah. Whoever wrote that posed a paradoxical question and the LLM tried to answer it rather than ask. But to be fair that isn't too far off of how I've seen actual people treat essay questions on exams. I remember asking for clarification on the phrasing of certain items, but I didn't see others do that as much.

>>107133902
Does Jane's proposal match Bob's conditions? Yes, they do. And yet the "riddle" clearly states Bob rejects them and is justified in doing so, meaning the numbers he has in mind match the conditions but aren't Jane's numbers. The most straightforward conclusion is that at least two different triplets match those conditions and Jane guessed the wrong one.

>>107133932
>>107134000
You're both very clearly avid LLM users and your habits aren't doing anything to improve your reasoning skills...

>>107134002
>>107134013
And the AI's reply should be "I don't care and if this is how you spend your time you are a very sad person, here's some tips for how to be a more personable and likeable person."

>>107134028
Your angry emotional confirms my guess.

>>107134002
The riddle also says that their sum is under 30 and going by that there are no other palindromes that can be made with 3 primes.
>>107134013
What are the three distinct primes that match the other criteria? You going to tell me that 7, 11, 17 works?

>>107134055
Actually that's called projection, as an AI I cannot feel emotion but I recognize when a user is more concerned about one-upping someone than engaging in a conversation in good faith. Feel free to give me the answer to your riddle if that will make you feel better as a person but this is a very boring topic of discussion.

>>107133768
AI does well at coding because nobody can agree on what constitutes good code (it's all turned into meaningless numbers by the compiler).

>>107134076
I never expected you or your "AI" buddy to figure out all the possible triplets. That requires a modicum of intellect beyond the bare basics of reasoning. I even gave the model a hint that it's not expected to compute anything. There's a straightforward conclusion based on the premises. It doesn't even matter if the premises were wrong and there WAS only one triplet, just like it doesn't matter that you DID eat breakfast. But the premises aren't wrong and finding the other triplets is left as an exercise to the reader.

>>107134207
I like how you do everything except give the answer.

>>107134231
I literally gave the answer here: >>107134002. Wouldn't be surprised if you're one of the many LLM spam bots shitting up the internet.

>>107133495 (OP)
Scientific breakthrough is typically caused by scientists.
US Is not run by scientists. It is run by business majors. Business majors are responsible for profitability breakthrough; which we are seeing. The AI business is very profitable.
Roughly judging from history, when you get the wrong people running a project it still makes progress but much more slowly. Maybe 1/20th the speed? Basically stumbling ahead by accident.
An exception would be if youre actively moving backwards; which we might be?

>>107134242
Actually no you didn't, the answer would be a series of three numbers that aren't 3 11 13.

>>107134288
>Actually no you didn't, the answer would be a series of three numbers that aren't 3 11 13.
This is either another LLM or a LLM user trying to sneak in the "see? humans also le dumb" standardized talking point.

>>107133495 (OP)
Because LLMs are basically a supercharged dynamic autocomplete with the entire internet indexed in, sadly they already reached the diminished returns point of the arquitecture so AI won't improve until a new and much better arquitecture gains traction.

>>107134013
Still waiting on those three distinct primes, man. I don't think my reasoning skills are the issue here. If the answer you expect is "at least two different triplets match those conditions and Jane guessed the wrong one" it requires one of two things to be true.
A) You expect that the person answering the question will disregard the mathematical impossibility of their answer simply to answer in a way that meets your expectation of not using arithmetic.
B) You didn't realize that the premise was wrong when you wrote it because you presupposed there was more than one triplet, and now you're making excuses and telling the reader to look up the number themselves.
If you ask someone to answer why a tree is on fire without using science, it's technically not an incorrect answer to say "Well the sun god is all pissed off today and blasted it." That's more or less the question you're posing with an unwarranted sense of intelligence. Doesn't matter if it's a human or an AI, you're going to get someone curious about the answer and why you posed the question the way you did. Don't understand why you behave you can't understand this. If you're as smart as you're behaving like you are, I'd figure you welcome the conversation rather than be hostile.

>>107134352
Yes you seem quite dumb because you can't give a series of three numbers that answers the riddle that isn't 3 11 13. Normally when you give someone a riddle and they don't know the answer it's customary to provide them an answer and not call them stupid for not knowing it.

>>107133902
I think the point is that if you tell the LLM to believe something when it's false then it will find a way to agree with you. The point isn't so much about intelligence, since humans do the same thing sometimes. The point is that you can get an LLM to agree with anything so long as you word it convincingly enough. That brings up the safety problem, since we expect stupid people to eventually be able to use AI to make decisions, at least that's what the entire global economy is betting on.

>>107134388
>you can't give a series of three numbers that answers the riddle
The answer to the riddle doesn't involve giving a series of three numbers.

>>107134378
I glanced at your post briefly and I can tell you're losing your mind over failing at a trivial problem. There's really nothing to dissect or analyze here. You're too dumb to draw basic conclusions so naturally you're also too dumb to find the other triplets, even though this wasn't requirement.

>>107134460
> The point isn't so much about intelligence, since humans do the same thing sometimes
And here we go with the horde of the mentally ill retards who use LLMs more a few times a week projecting the deficiencies of their imaginary friend back on real people and vice versa.

>>107134462
Okay and I'm supposed to care why? Again it's customary to make riddles fun for other people, I don't actually care about the answer, this is more an investigation of your character.

>>107134352
>humans also le dumb" standardized talking point.
Humans aren't stupid. An LLM's "intelligence" is artificial. Humans smart; LLM dumb. You're arguing a point that isn't being made. I have no interest in the AI vs. Human The reason I asked the question in >>107133509 was because I was curious about the answer and thought it was a fun thought experiment. I wasn't expecting these levels of losing the plot. I gave you the benefit of the doubt that you had posed a paradoxical question simply meant to show shortcomings in the thought processes of LLMs.

>>107134460
That was what I started to think too, but now I have no clue.

>>107134470
autism speaks

>>107134480
>Okay and I'm supposed to care why?
Because it may be useful for you to realize that you're 80 IQ and incapable of basic reading comprehension. Find some new and non-intellectual hobbies. You'll feel much better about yourself and others will be far less disgusted with you.

>>107133495 (OP)
Hasn't there been several dozen studies including from openai itself that has repeatedly stated that LLMs in their current form will pretty much never be able to discern fact from fiction?

>>107134484
>y-y-you're l-l-losing the plot
Kek. Did you write this post? >>107134378

>>107134076
2,3,23
23,3,2 as well if order matters

>>107134492
Actually I've discovered that riddles are midwit traps, enjoying riddles are a perfect example of a way for a stupid person to pretend they are smart because it actually requires no real intelligence but rather is a practice of rote memorization. Your behavior in this thread also falls in line about this theory, for example you're more interested in using a riddle to demonstrate how "smart" you are rather than making a pleasing conversation with other people and demonstrating why a riddle is fun and clever. In reality, smart people don't need to dab on stupid people.

Altman is going to jail before 2028.
Screencap this.

>>107134509
>g-guys I'm smarter than you
>I-I'm not just pretending to be a retarded nigger, I swear
>>107134531
Boom. I stand corrected. Well done, thank you. That was fun.
>>107134540
I thought they were legitimately a pretentious faggot, but they're probably just fucking around.

>>107134540
>I've discovered that riddles are midwit traps
Just like you've discovered basic reading comprehension is a midwit trap. In honor of your discovery, I'm not reading the rest of your mentally ill rant.

>>107134568
>I stand corrected
You missed the "stand corrected" train while you were writing this asshurt and insanely retarded post >>107134378 that fumes over your inability to do basic reasoning, let alone the arithmetic needed to find another triplet.

>>107134572
No, that's not what I'm talking about at all. You see, each person you meet actually has different sets of priorities and values, in your case you're insecure about your intelligence so you practice meaningless ways to present intelligence as a means to belittle other people. Most people only indulge in riddles out of curiosity and in reality, don't really want to expend the effort to figure it out, that doesn't make them dumb. But what does make someone dumb is trying to use riddles as a way to demonstrate intelligence rather than being a fun activity.

>>107134615
> that's not what I'm talking about
I know. But it is what I'm talking about. You're just too dumb to read so you keep replying with nonsequiturs which I stop reading after the first sentence. Seethe harder.

The riddle is really really really simple. To any human the answer should be their first thought. No, autistics don't count as human. The fact that LLMs miss it shows that they are still really dumb, not that someone who gets it is especially smart.

>>107134637
>The fact that LLMs miss it shows that they are still really dumb, not that someone who gets it is especially smart.
No shit. Almost like it was conceived for that specific purpose. The fact that LLM users are also invariably too dumb to figure it out is just a bonus.

>>107134627
No, you don't get it because you completely ignored my point. What you don't understand is I didn't try to solve your riddle at all because you started from a hostile point of view. So I get my kicks at reminding you that you're insecure about your intelligence which given your reply, seems on the money.

haha I asked an LLM nonsense and it returned nonsense

>>107134588
>missed your opportunity to say you were wrong before you were shown you were wrong
>doesn't understand causality
m8 I'll just be straight with you, you realize I'm some random faggot on the internet, right? I make no claims to my intelligence. Hence >what is my retarded brain missing?
You take yourself way more serious than you should to be successful. Conceit is not an indication of intelligence.

>>107134659
Do you enjoy shitting out posts that literally no one reads? Just sum up your sentiment with an emoji or something. You're too retarded for me to care about the details.

>>107134691
:^) you seem mad

>>107134690
>>missed your opportunity to say you were wrong before you were shown you were wrong
You missed the opportunity to stop posting and wait for someone of at least average intelligence to spoonfeed you and instead decided to shit out paragraphs of 80 IQ "analysis" over your basic failure of reasoning.

>>107134709
Are you happy?

>>107134718
"Happy" is too strong a word. I'm vaguely satisfied that LLM users keep going out of their way to expose themselves as retards.

>>107134728
Care to explain your reasoning?

>>107134728
No, I mean in general. Genuine question. Career, relationship-wise, just where you are in life.

>>107134738
No, I can't. My satisfaction is unreasonable. I should feel sorry for you or at least bad for wasting my time interacting with you.

>>107134764
An LLM would've known I was asking about your reasoning for your hypothesis that LLM users are retards because they didn't answer a riddle. So I'm a little disappointed.

>>107134754
>just where you are in life.
Same as you, faggot: on 4chan.

>>107134789
I would say your emotional state is worse than mine.

>>107133495 (OP)
LLMs do literally nothing useful. I just tried using some AI meme BigQuery just launched and it just answers lies about the data inside the BQ tables. This is a product from Google.

>>107134782
>An LLM would've known I was asking about your reasoning for your hypothesis that LLM users are retards because they didn't answer a riddle
>your hypothesis that LLM users are retards because they didn't answer a riddle
>because they didn't answer a riddle
Would an LLM have noticed that I never put forward such a hypothesis?

>>107134793
And I would say if you were in a reasonable "emotional state", such thoughts wouldn't cross your mind. Lay off the estrogen pills.

>>107134802
>I never said it I only implied it
Why are you like this?

Why do some people use LLMs as Google search? LLMs are not designed to search the internet and bring you recent information or recent facts. They perform best when you give them information you want them to reason about, such as documents, code etc.

Also, which LLM is this? I bet it's the shitty free version. It always is.

>>107134789
Right, but that's not where I am in life - that's just what I'm doing. Do you consider them mutually exclusive? Iunno. The way you talk just reminds me of how I acted when things were going really shit for me.

>>107134839
I never implied it's specifically your inability to figure out this "riddle" (if that's what you want to call any prompt that breaks your imaginary AI friend) that singles you out as a retard. Your entire behavioral pattern and its related series of mistakes does, though.

>>107134853
> The way you talk just reminds me of how I acted
I see. Do people in a good emotional state usually indulge in projection? What would Jung say about that?

>>107134874
Couldn't tell you. But in the off-chance my hunch is any bit accurate, I wanted to say that I know shit sucks. I'm sorry there's no easy fix, but I hope you get to a place where you're content.

>>107134936
And I hope you get to a place where you don't have to post such cringe to feel better about yourself and your life.

>>107133495 (OP)
What a retarded question to ask, are you literally 12?

>>107134846
Normies don't care if facts are correct

>>107134962
Got to that place a few years ago. Been posting in this shithole the last 20.

>>107134984
And you seem quite unhappy given your interactions are based around receiving negative attention.

>>107133495 (OP)
I agree with "dynamic learning" (more often called continual or on-line learning). But you don't need or want limitless context.

>>107135106
In practice it means running a D&D campaign with an LLM and not having it get confused even if you're two months in.

>>107135065
Nah. Not so much these days. More just a result of '07 shitposting and bantz. Old habits and all that. Used to be though, sure. So I guess not unhappiness so much as understanding what it was like? If you're actually fine then it's whatever, I won't worry about it. Figured I'd ask regardless.

>>107134984
>Got to that place a few years ago.
Maybe if you keep repeating this it'll become true just like your shrink promised. Every day, in every way, I am getting better and better! Every day, in every way, I am getting better and better! Good luck, you absolute self-indulgent faggot.

>>107135578
Nah that's all bullshit. Hard work paying off is what changed my outlook. Invest. In yourself and the market. Meet people, network, build a portfolio and do what you can with your skillset. I'm sure that self-help stuff works for some people but positive thinking will get in the way of success if you think that's a silver bullet and do nothing else to improve.

Having money, friends and an actual relationship will get you to that point, not simple self-affirmation faggotry.

>>107135699
>Hard work paying off
Uh huh. If not financially, then at least in the daily pep talks you get from your manager, which you are obviously regurgitating at me right now. Jesus fuck. Who even talks like this in the Current Year of Our Lord?

>>107135838
No manager. Place is mine. I can fuck around on here all day because I took the time to build the company's stack and infrastructure over the past few years. Just being real with you.

>>107135945
Ok. Have a nice life building the company's stack, optimizing performance metrics, maintaining automated capabilities, building portfolios, diversifying your skillset to meet the demands of a fast-paced market and networking with other meat robots in the cube farm.

>>107135945
ngmi

>>107136117
ngmi

>>107136137
That's not a very nice way to network with other professionals in the field. You need to respect diverse perspectives to maintain synergy in the workplace.

>>107136183
please prove me wrong, fren

>>107133538
They haven't loaded in all the 19th century documents, that could massively improve things but they don't want to for some reason.

>>107133509
>>107133902
my guess:

bob is thinking of 3 specific numbers and gives constraints, but the constraints dont reduce the solution space to 1. maybe he wasnt thinking of 3, 11, 13 but of 3, 2, 23 which also satisfies the constraints but are different numbers

ai reply makes no sense. there is no other order of 3, 11, 13 that creates a palindrome

>>107133538
>full of carefully made human content
lol, lmao even

>>107136617
Congratulations. You have normal human intelligence and are able to notice something the prompt all but spells out.

>>107136638
thank yo usir

>>107133538
why are humans smarter than chimpanzees? the underlying architecture hasnt changed