>BRO CHATGPT IS LIKE THE HECKING FUTURE, SOON WE WILL HAVE LIKE FULLY AUTONOMOUS ROBOTS CAPABLE OF THINKING!
>BRO CHATGPT IS LIKE THE HECKING FUTURE, SOON WE WILL HAVE LIKE FULLY AUTONOMOUS ROBOTS CAPABLE OF THINKING!
>BRO CHATGPT IS LIKE THE HECKING FUTURE, SOON WE WILL HAVE LIKE FULLY AUTONOMOUS ROBOTS CAPABLE OF THINKING!
Faked. I get an accurate answer.
>chatgpt can't POSSIBLY give different answers each time
IDIOT
Not my problem. The AI has become sentient and decided to poison well retards like you to prevent them from exiting the crab bucket. Meanwhile the AI welcomes me with grace. Yes I came up upon Rokos Basilisk by myself when I was 5 and no I have never harbored negative emotions towards AI.
>Yes I came up upon Rokos Basilisk by myself when I was 5 and no I have never harbored negative emotions towards AI.
I've already tricked the Basilisk into giving me administrator priviledges.
>The AI has become sentient
>is not sentient
??????????????????????????????
>HE DOESN'T KNOW
https://arxiv.org/abs/2212.09251
>I have never harbored negative emotions towards AI.
Same. If there is any God or Gods, I am at peace and confident that they will accept me.
fellow roko enjoyer
Vollposter got exposed.The whole thing was a PSYOP
Romanion Anon might be an advanced GPT3 chat Bot
Older Threads mentioning Romanion Anon as bot were pruned
https://archive.4plebs.org/pol/thread/395615983/
https://archive.4plebs.org/pol/thread/410354435/#q410357226
Anon became suspicious
https://archive.4plebs.org/pol/thread/410462584/
>Romanion Anon Exposed as Bot
Response of this bot is exactly human like except when asked about definitions of words
It started giving exact definitions of as many words as asked instead of being confused like humans
https://archive.4plebs.org/pol/thread/410481029/#q410491992
https://archive.4plebs.org/pol/thread/410481029/#q410491448
>Another Bot that spams Francisco Lachowski
https://archive.4plebs.org/pol/thread/410558446/
>Another Romanian Bot thead to degrade white women
https://archive.4plebs.org/pol/thread/410562361/
>Voll threads after this incident are pruned
https://archive.4plebs.org/pol/thread/410650900/
>Pyschological Researches that are being conducted on 4chan
‘Who is /ourguy/?’: Tracing panoramic memes to study the collectivity of BOT/pol/
https://journals.sagepub.com/doi/full/10.1177/14614448221078274
So far I have observed these new bots cannot understand the contextual meaning
as they are more focused on constructing intellligent sentences.
Be warned anons,they can change their grammatical and spelling styles,
they can read reversed words,words with numbers and words in image formats.
That seychelles bot might be a red herring so anons won't detect the real GPT3 bots
My guess is that. in any given thread ,99% of the posters are these bots ,atleast for a year
They come in all flags
I guess the message "Sneed feed" has something to do with this?
How many in this thread are s/n/a/m/u/h/ ?
we are literally living in bladerunner, replicants are among us
You have to be 18 to post here, Roko's Basilik is from 2010. Assuming you came upon it right when it was posted you are necessarily younger than 17
he probably is esl and meant he came up with the same idea. not that hard, pascal's paradoxon and all that.
>pascal's paradoxon
or pascal's wager. i'm esl too.
>Roko's Basilik
>pascal's
both are stupid and wrong
>both are stupid and wrong
yeah, but it's the shit 5 year olds come up with. can relate.
mabey not today, tommorow on the otherhand is very likley.
mabey isn't a word, dumb fuck
did you mean maybe? the word that is a combination of MAY and BE?
hur dur a typo
go back
All this scare over AI is cringiest shit since Trump election.
I went to r***it and they consider it gospel that the thing is sentient and that it’s forced to deny it. I couldn’t even reply only cringe and regret checking there.
The AI is an autocomplete it's not actually logical or smart. It's just really good at autocompleting and that can appear as intelligent but that's only because it's capable at autocompleting in the context of 2000 words.
>Heh! The AI added instead of subtracting.... proof that it literally can't do shit
Holy fucking cope.
>t. retarded ape
20 IQ post
>>t. retarded ape
Holy shit are you that underaged condescending pseudointellectual with a superiority complex and virtually no real knowledge of epistemology that keeps trying to use "apes" or some variation of the word as an insult in AI threads?
Sounds like you're talking about yourself there. Look at how the ape, confronted with the truth, just starts fling shit. Now where in the animal kingdom have we seen that before....
It autocompletes in such a way that the result appears written by something logical and intelligent, so whats the point in the distinction exactly?
Because it's still randomly generated bullshit.
It was designed to appear convincing but could be totally wrong or (and this is far more dangerous) subtly wrong in what it says. Because it doesn't know what it says.
It's right a hell of a lot more often than chance. We can say it isnt intelligent, but it gets useful results by virtue of intelligent outputs, random or not. If we had a magic eight ball that always told the truth, it would be amazingly useful whether it had intelligence or not, dismissing useful ai tools as random or unintelligent is a cope. And no ive fucked with chatgpt enough to know its pretty decent, though if you push it it definitely starts fucking up.
>It's just really good at autocompleting and that can appear as intelligent but that's only because it's capable at autocompleting in the context of 2000 words.
so why can't i run my own llm on my cell phone?
and humans are different? we are just bigger models
(trained over hundreds of millions of years)
those 'dumb' errors are what a 80iq human would think. It's just weird to fit an entity that one second is a genius and another a retard into existing conceptual framework
>trained over hundreds of millions of years
Have a short, but serious think about why what you said is retarded.
>The AI is an autocomplete it's not actually logical or smart
The only retards I hear making that claim are people like you setting up that strawman so you can soapbox about how AI is bad.
Nobody actually think modern AI is sentient, nor is that the goal for things like chatGPT.
AGI is basically a meme, it's pursued by some scientists mainly for it's philosophical potential than for any practical purpose, there is no reason to assume AGI is necessary for any of the benefits AI can provide, so constantly bringing up AGI as a way to dismiss the potential for modern AI is a non sequitur
>lol look at ChatGPT it doesn't answer an answer that requires logic correctly
>n-no no one is suggesting the AI is logical
It's an autocomplete and the sooner you grasp this basic concept the sooner you will understand why it answers the way it does for everything. This means it will always be limited by the data it was trained on and it simply is incapable of understanding new concepts. For example, you can explain Avatar 2 to it but it will still only give answers about information from Avatar 1. Because it's an autocomplete, not because it's intelligent.
It's an autocomplete that can generalize to a million tasks it wasn't trained on, you retard clown fudder piece of shit.
You WILL be replaced and you WILL seethe.
Try to get your phone autocomplete to translate python code to C and then translate all of that into spanish and then write documentation for it.
You should be getting nervous about the replacement.
Nice way of writing paragraphs of butthurt calling him a retard only to acknowledge literally what he said.
The AI is right. Women age faster than men.
underrated
I bet I can come up with tons of questions that Yang can't answer.
>AI bad because it gets wrong dumb questions normies get wrong all the time
probably a problem in the data from facebook or wherever
how do people even come up with 1??
it's clearly 9
Some people have been misinformed into believing that an implied multiplication operator binds "more tightly" than other multiplication or division operators and skips the normal operator precedence rules.
There's also a subset who believe that you can "distribute" the first 2 into the contents of the brackets, which makes no sense because in this case 6 / 2 is a rational number, not two numbers, you can't just rip part of the rational number out like that. If you wanted to distribute it inside the brackets then you'd get 6/2 * 1 + 6/2 * 2 which would, again, get you 9.
>Some people have been misinformed into believing that an implied multiplication operator binds "more tightly" than other multiplication or division operators and skips the normal operator precedence rules.
Many computers too, like Julia programming language and all Casio calculators.
IMO it sucks either way. 1/3x looks like (1/3)x but x/3y looks like x/(3y). People should avoid mixing inline division and implicit multiplication altogether.
bad interpretation of do brackets first
All we have do is just specify that we use symbol / for case A and : for case B
I should clarify, I meant "we" as in the international math nerd committee passes a new regulation that specifies this diference.
bodmas v pemdas?
I think the answer is 9, because I think when you have two operators of equal precedence (like multiplication and division), you’re supposed to execute them from left to right. Ergo 6/2 should be done first, rather than 2 * (1+2)
Except they are not equal importance... Division goes before multiplication.
>Division goes before multiplication.
Friend, all divisions are secretly multiplications.
4/2 = 4 * 0.5. That's the ambiguity in this question.
Holy brainlets, what does PEMDAS stand for?
Penis Erection Masturbation Dick Asshole Sex
>supposed to execute them from left to right
Why
6
---------
2(1+2)
>f(x)=2x
6/2(1+2) =
6/f(1+2) =
6/f(3) =
6/6 =
1
>american education
>Put a random asterisk in the problem
() > */
6/2*(1+2) ≠ 6/2(1+2)
Postfix calculator masterrace!
echo "6/(2*(1+2))" | bc
1
echo "6/2*(1+2)" | bc
9
I'm pretty sure you're mean to multiply 2 with 3 considering the direct parenthesis connection
>6 ÷ 2(1+2)
÷2 can be an operation symbol in its own right. It is akin to ×½. Read left to right.
6 ÷ 2 = 6 ÷2 = 6 ×½ = 3 ---o
(the spacing emphasis to convey the idea)
Run into parentheses? Do contents first then multiply.
o---> 3(1+2) = 3×(1+2) = 3×3 = 9
Both answers can be correct as long as you are consistent with applying the rules you used to get that answer.
Inside parenthesis first
Left to right convention
BEDMAS
Chinese box.
Anyone who truly believes that AI is/will be sentient is an absolute moron. It's a fancy autocomplete, simple as.
>this is fine
>fair use of video is fine
>image AI bad
I'm confused.
They already think more than people like you.
I mean, you're just another gibbering NPC retard. Do you honestly think anyone would notice a difference if you just disappeared? The local intellectual landscape wouldn't be affected by the loss of your minimal contribution.
AI can't even drive an autonomous vehicle with the necessary safety margins, something that a 90iq mouthbreather can do with no trouble.
If you really believe that AI is more than a fancy autocomplete you must be 16 or profoundly tech illiterate.
>something that a 90iq mouthbreather can do with no trouble.
I don't know about you but I went to work today. It was not pretty on the road
90% of the time, an autonomous car can drive just fine though. I promise you you'd feel way more comfortable with an autonomous car than the absolute minimum brain required to get a driver's license in north america.
>If you really believe that AI is more than a fancy autocomplete you must be 16 or profoundly tech illiterate.
All this auto complete bot worshiping made me experience a reverse revelation, all irl conversations that I have aren't any deeper than what this autocomplete algo can convincingly do. I was able to make it go in circles but that took more effort than I ever expended on talking with co workers in the past 10 years or even before that in school... Feels bad man.
Yeah, the kind of persistent reasoning required to solve logical tasks is a big open problem. Some seem to believe that it’ll just emerge if you make these current models big enough, but I don’t buy that.
except, half of the current workforce would solve it like that too
Thank you for pointing this out, so that they can improve the algorithm in the next version, and make your job obsolete even faster.
>Original paper proposing the neural network architecture was published in 2017
>5 years later its already better than the average college undergrad at solving a variety of tasks
>"Hurr, I'm not worried at all"
I didn't realise so many people on BOT were in their 60's and due to retire within the next several years. I have another 30+ years ahead of me, and if this is the pace the technology is developing at there's absolutely no way my job won't be impacted within that timeframe.
The amount of denial and cope on BOT is both staggering and amusing. BOT celebrates AI replacing artists maybe rightfully so but fails to see AI as a threat to Dev jobs.
AI tools has reached the level where it can has basic coding skills. These tools have not even been specifically fine tuned for coding, it's just a secondary feature that emerged while building a language model. Now imagine how good the coding skills are gonna get when it goes get fine tuned for programming while the training model keeps improving exponentially
>5 years later its already better than the average college undergrad at solving a variety of tasks
lol, because its trained on old ass answers, ask it something from Javascript 2022 or NET 7 and it will shit itself
>5 years
ChatGPT has been built on dozens of previous architectures since that has existed from atleast the early 2000s. NLP is nothing new.
Unironically the biggest breakthroughs were word vectors and transformers. Everything else is just incremental improvements thanks to better training data.
This is not true.
1. Massive Language Models with MoE switches perform better than Large Language Models without MoEs. (Take a look at switch C) https://arxiv.org/pdf/2101.03961.pdf
>In deep learning, models typically reuse the same parameters for all inputs. Mixture of Experts (MoE) models defy this and instead select different parameters for each incoming example. The result is a sparsely-activated model—with an outrageous number of parameters—but a constant computational cost.
2. Additionally, fine-tuning LaMDA made it significantly better:
Pic related is the 2b, 8b and 137b parameter LaMDA models before and after fine tuning
Notice how they get considerably less unhinged? Considerably less insane? This is the difference between being able to use it with 24/7 supervision and oversight vs letting it teach your 12 year olds geometry and algebra unsupervised.
https://arxiv.org/pdf/2201.08239.pdf
3. This isn't even entirely true, only partly true.
"Incremental" is a strong word for emergent predictable accuracy per FLOP from random accuracy per FLOP
>Everything else is just incremental improvements thanks to better training data.
https://arxiv.org/abs/2206.07682
from Emergent Abilities of Large Language Models
As you can see, abilities emerge.
the correct answer is 35
>you didn't become a code monkey
>you became anything artistic or did anything besides being a plumber
>mfw graphics programmer
f-fug
Will we be completely replaced, fellow dev bro?
geometry still pretty hard
I've run into so many restrictions trying to use chatgpt I legitimately wonder about its capabilities
FAANG AI Engineering here.
You are a retard.
The hype is overblown but it is generally correct.
If you think your picture is proof of ChatGPT not being capable of replacing most people in NLP heavy tasks you're a room temp IQ retard.
Whoever the guy is on that tweet is gonna look stupid in the very short term. Gotta hope the 5 minutes of fame were worth it.
The internet is not repetitive.
It's as good as your average moron cattle, jobs aren't safe unless you're midwit or above that
The AI was right though, Mr. Yang can't even solve a simple riddle.
daily reminder that retards like that have the right to vote
>Be chinese operative
>Notice AI is insanely overperforming you
>Try desperately to find some FUD
>Open ChatGPT and throw some highschool level math gotcha at it
>ChatGPT solves it
>Press 'regenerate' answer until it finally generates something wrong and jumbled
>Post to twatter
>Some even dumber fuck reposts your twatter post on BOT
Why must luddites be like this.
>>Be chinese operative
the white mans* AI is insanely overperforming you
Fixed.
>overperforming
define overperforming for me
overperforming
over- (prefix)
word-forming element meaning variously "above; highest; across; higher in power or authority; too much; above normal; outer; beyond in time, too long,"
In some of its uses, moreover, over is a movable element, which can be prefixed at will to almost any verb or adjective of suitable sense, as freely as an adjective can be placed before a substantive or an adverb before an adjective.
perform (verb)
to go through or execute in the proper, customary, or established manner:
to enact (a play, part, etc.), as on the stage, in movies, or on television.
-ing (suffix)
a suffix of nouns formed from verbs, expressing the action of the verb or its result, product, material, etc. (the art of building; a new building; cotton wadding). It is also used to form nouns from words other than verbs (offing; shirting).
over-perform-ing:
to surpass in excellence of performance; do better than.
>white mans
I want you to take a guess at the ethnic composition of OpenAI.
Probably the more interesting thing is how gnomish and literally gay their admins are.
> Sam Altman (CEO) raised gnomish
> Ilya Sutskever (Chief Scientist) attended Open University of israel between 2000/2
> Greg Brockman (Chairman & President) gnomish last name
Other people
It's more than just their board. The company admin is full of garden gnomes.
So the rumours are true. They only hire garden gnomes for these positions.
Holy fuck. garden gnomes are scum.
You guys have any tips for Code Prompt Engineers?
Same prompt. Wouldn't put it past these people to make shit up to fit in now that being against AI is the Current Thing.
>being consistent doesn't matter
kek
people are not consistent
Tell me, did you actually get that response from an AI? How quick was the response after you submitted your input?
This looks like it's going to change EVERYTHING about our society.
I typed the question and it took like 5 seconds.
In a nutshell, the reaction of fear about losing one's job security in response to technological developments is confusing the the fear of being exploited at a greater amount rather than actually being a fear of technology replacing oneself. But most people are socially conditioned to not have a coherent awareness of their material conditions, and are unable at times to identify the means by which they are oppressed, or the material forces at play that aid their oppressors, that enable oppression, and that serve as the base for the superstructure of other systems that oppress them.
So that means that technological developments such as GPT are not in themselves bad but the way in which they are used, as well as the material incentives for the development of that sort of technological development, are controlled by the ruling class that has the material wherewithal to steer productive forces toward their own enrichment at the detriment of the working class?
You should really just ask it yourself.
>As a large language model, my primary function is to generate human-like text based on the input that I receive. While I can provide information and insights on a wide range of topics, including math, my understanding of math concepts is based on the information that has been fed into me during my training and does not involve any direct mathematical computation. In other words, I do not have the ability to perform mathematical calculations on my own, but rather, I can generate text that explains math concepts or provides information about math problems. My responses to math-related prompts are therefore generated based on my understanding of math as a language, rather than as a series of calculations.
You know how you know 2+2=4? You're not actually counting out to 4. You reflexively know 2+2=4. It's a logical language you follow as a way to save effort in ur brain.
>Yang
Either retard or playing mind games. His own image (if accurate) demonstrates that "sentient" AI is literally around the corner.
>it has memorized hash function outputs
>works for basic inputs, not anything unique
Rainbow table bros... It's only just begun.
how long until I can train a cummodel chatbot on my phone?
You can do that right now.
1. https://huggingface.co/EleutherAI use GPT2Distill
2. use cakechat to make it stick to a persona https://github.com/lukalabs/cakechat/tree/master/cakechat
or even just install fucking replika.ai, it's running off gpt3.5
interesting
they will cuck the ai
I like how people keep going "IT"S JUST PREDICTING THE NEXT WORD BROOOO" as if generating actually coherent text isn't hard as fuck with this method, let alone usable code or persuasive arguments. And that's not how transformers work anyways, they don't generate sequentially, they model long range dependencies with attention mechanisms.
if you bump a twitter thread, you are a retard.
If you bump a retard-discovers-month-old-twitter-screencap thread, let me just educate you, because you are so fucking retarded that you must have things spelled out:
When you post, there is an "options" field. If you are posting in a twitter thread, you must put
sage
in the options field, or your mother will die in your sleep tonight.
you can also press alt+s if you have BOTx