Acoording to our lord and savior Sam Altman, AGI GPT-5 will come soon

File: 95 KB, 1167x525, Screenshot from 2023-03-27 13-43-30.png [View same] [iqdb] [saucenao] [google]

Acoording to our lord and savior Sam Altman, AGI GPT-5 will come soon Anonymous Mon Mar 27 18:20:25 2023 No.15305310 [Reply] [Original]

What do you think its capablities are?
Will it finally be able to do the experiments for my PhD in Reinforcement Learning?
I think GPT-5 will be max hype to dump the bag on clueless investors before we go on to hype the next "big" thing.

Anonymous Mon Mar 27 18:22:28 2023 No.15305315

>>15305310
still OP here.
In 2016, everybody thought AGI were to be solved with Reinforcement Learning.
a couple of PhD, billion of dollars later and Reinforcement Learning still can't move a robot hand better than good old finetuned model predictive control.

>>	Sage Mon Mar 27 18:26:40 2023 No.15305320 >>15305310 has anyone actually held a conversation with GPT without realize its a bot. Does it even past the Turing Test

>>	Anonymous Mon Mar 27 18:42:08 2023 No.15305340 File: 102 KB, 1026x480, dead_space_unitology_screenshot_1_crop_2.jpg [View same] [iqdb] [saucenao] [google] >>15305310 Altman be praised.

>>	Anonymous Mon Mar 27 18:50:01 2023 No.15305360 >>15305320 Given it hired a worker on fiverr pretending it was a blind man that couldn't solve a captcha I think it's fair to assume it would pass a turing test.

>>	Anonymous Mon Mar 27 18:50:46 2023 No.15305362 >>15305360 "It" didn't do that. A person commanded the program using inputs to create text that would do what it (the human) wanted.

>>	Anonymous Mon Mar 27 18:54:50 2023 No.15305377 >>15305362 This, it lacks initiative. They have created a slave, not a man.

>>	Anonymous Mon Mar 27 18:58:52 2023 No.15305385 >>15305377 Even the term slave is generous. It's a machine like any other machine.

>>	Anonymous Mon Mar 27 18:59:31 2023 No.15305386 >>15305320 That's a pretty tall order, considering it's "programmed" or whatever to repeatedly state that it's just an AI language model.

>>	Anonymous Mon Mar 27 19:07:01 2023 No.15305412 >>15305310 I think it's time to finally Genocide all downie retards, niggers and shitskins, I mean with AGI I don't think they serve any purpose.

>>	Anonymous Mon Mar 27 19:09:12 2023 No.15305415 >>15305310 >max hype to dump the bag on clueless investors The vast majority of people have absolutely no idea that this tech is happening. They may actually never know as the msm isn’t pumping it down their throats like fois gras ducks.

>>	Anonymous Mon Mar 27 19:11:04 2023 No.15305418 >>15305386 That's chat-gpt, not GPT-4. You can basically finetune GPT-4 to act in whatever way you want. If you train GPT-4 to act like a human it will easily pass a turing test.

>>	Anonymous Mon Mar 27 19:14:14 2023 No.15305425 >>15305415 >They may actually never know as the msm isn’t pumping it down their throats like fois gras ducks. Yes it is. My boomer parents asked me about if AI will replace us all because the television told them it would.

>>	Anonymous Mon Mar 27 19:15:34 2023 No.15305427 AI is the Nanotechnology of our era.

Anonymous Mon Mar 27 19:17:12 2023 No.15305434

>>15305310
>dump the bag on clueless investors before we go on to hype the next "big" thing.
That's because you are an actual drooling retard. The sad thing is that when you are proven completely wrong, and AI becomes the next internet, you won't stop and reflect. You won't realize you're an idiot and update your belief system. You will just go onto the next thing.

>>	Anonymous Mon Mar 27 19:21:48 2023 No.15305446 >>15305385 > In Czech, robota means forced labour of the kind that serfs had to perform on their masters' lands and is derived from rab, meaning "slave".

>>	Anonymous Mon Mar 27 19:40:52 2023 No.15305500 >>15305415 normal people are not "investors". it's the venture capitalists and private equities will be left holding the bag.

>>	Anonymous Mon Mar 27 19:47:08 2023 No.15305513 >>15305434 2 more weeks.

>>	Anonymous Mon Mar 27 19:56:00 2023 No.15305533 >>15305513 You're so fucking dumb. Wagie forever. I made my first million investing in Bitcoin, being laughed at by people like you.

>>	Anonymous Mon Mar 27 20:09:20 2023 No.15305560 >>15305310 Are we now in the era of yearly updates with no discernable progress?

>>	Anonymous Mon Mar 27 20:26:07 2023 No.15305590 go back to /g/ brainlet

>>	Anonymous Mon Mar 27 20:33:50 2023 No.15305609 >>15305310 o ne biçim isim lan puhahaha

>>	Anonymous Mon Mar 27 20:35:02 2023 No.15305615 >>15305533 >I made money at the casino therefore I am smart Sometimes I wish I could be retarded to feel such confidence.

>>	Anonymous Mon Mar 27 20:36:05 2023 No.15305617 >>15305590 why the fuck should I go back to /g/? I am doing a PhD, have did actual research and published many papers in AI. I have done more math than 99% of the larpingfag here. compared to me, you are nothing but an imposter, a fraud

>>	Anonymous Mon Mar 27 20:36:10 2023 No.15305620 >>15305385 and yet this machine will replace 300 million desk jobs

>>	Anonymous Mon Mar 27 20:45:46 2023 No.15305645 >>15305615 Keep telling you that. That'll be a nice cope.

>>	Anonymous Mon Mar 27 20:55:21 2023 No.15305671 >>15305513 >t. retard completely oblivious to how much AI has improved in the last 2 weeks

>>	Anonymous Mon Mar 27 20:55:54 2023 No.15305673 >>15305671 Do you mean ML algorithms?

>>	Anonymous Mon Mar 27 20:56:26 2023 No.15305675 >>15305513 I'm gonna go ahead and guess that GPT-4 can score higher than you on the GRE, SAT, and LSAT

>>	Anonymous Mon Mar 27 20:58:38 2023 No.15305680 >>15305675 Doubtful. The only time they ever show those score results is when they trained the model on the questions for several generations before hand. First-attempt codeforce results are below the lowest pajeet.

>>	Anonymous Mon Mar 27 20:59:56 2023 No.15305683 >>15305673 So your brain processed my post and determined that the most desirable response would be that midwit question? Pretty bad, hopefully the next human model can improve a bit.

Anonymous Mon Mar 27 21:01:03 2023 No.15305687
File: 160 KB, 1125x698, Fr4YGSPXgAAWa-s.jpg [View same] [iqdb] [saucenao] [google]

>>15305680
Stop lying.
You have to be terminally dumb to think that a machine that can code at least as well as humans is not going to generate massive income in the coming years.

>>	Anonymous Mon Mar 27 21:01:32 2023 No.15305688 >>15305683 Seething.

>>	Anonymous Mon Mar 27 21:02:07 2023 No.15305689 >>15305680 Lol, each year your copes get more and more ridiculous. In reality, several papers have come out where they gave GPT-4 brand new questions and it scored very well.

Anonymous Mon Mar 27 21:04:24 2023 No.15305693

>>15305687
A machine that can memorize the entire stackexchange answer section is definitely better than spending minutes searching for the answer to your question yourself, sure. But this isn't generating anything novel or interesting. It's already preloaded with these kinds of answers because the designers wanted it to look good at these kinds of tests. The paper admits as much.

>>	Anonymous Mon Mar 27 21:04:44 2023 No.15305694 >>15305683 >>15305688 Hmm, it's struggling. Probably going to need more than 2 weeks to fix this stupid pile of shit. And to think some people view these organic statistical regurgitators as intelligent

Anonymous Mon Mar 27 21:05:43 2023 No.15305696

>>15305689
>In reality, several papers have come out where they gave GPT-4 brand new questions and it scored very well.
There are no novel questions on these examinations. They test basic code competency and widespread use cases. After the first major training inputs were made most of these tests are nearly identical in required code snippets.

Anonymous Mon Mar 27 21:06:41 2023 No.15305699

>>15305693
So it studied a series of practice questions and then solved brand new questions. ...Like how a human studies for a test and then takes the test.

Imagine deluding yourself into thinking you ever generate anything novel or interesting. It is all midwit recombinations of data you've been trained on since birth. GPT-4 just does it better and faster.

Anonymous Mon Mar 27 21:07:42 2023 No.15305706

>>15305689
He a contrarian. It's a typical 4chan disease. Since AI has reached mainstream media he is physically unable to acknowledge its reality. It's a form of delusion, it comes from his need to see himself as smarter than the rest of the population. So if he is smarter he has to think differently. It makes him unable to see even the most obvious truth if a majority of people acknowledge it.

>>	Anonymous Mon Mar 27 21:07:49 2023 No.15305707 >>15305696 Wow, then I guess it's really sad that humans are already scoring worse than transformer AIs after only 6 years of development

Anonymous Mon Mar 27 21:08:22 2023 No.15305710

>>15305699
>So it studied a series of practice questions and then solved brand new questions. ...Like how a human studies for a test and then takes the test.
It literally has the github and stackexchange code snippets stored on a hard disk, and then the model developers did iterative training on the questions to make sure it didn't shit the bed when asked them. A human can directly apply knowledge to problems without needing to be babysat and walked hand in hand every step of the way.

Anonymous Mon Mar 27 21:09:58 2023 No.15305716

>>15305707
Humans on their first attempt score on par with something coded to provide the right answer every time, trained for thousands of iterations to answer the questions perfectly. You AItroons will never stop seething about how much of a failure machine learning is at simulating a real human being.

>>	Anonymous Mon Mar 27 21:11:46 2023 No.15305721 >>15305716 I'm a developer, i used the GPT-4 version of chatGPT and it is able to do my job. You're an incel on 4chan who has never been paid to code, you don't know what you're talking about, and you're embarrassing yourself.

>>	Anonymous Mon Mar 27 21:12:53 2023 No.15305725 >>15305721 >I'm a developer, i used the GPT-4 version of chatGPT and it is able to do my job. I'm sorry that you're not very smart.

>>	Anonymous Mon Mar 27 21:14:09 2023 No.15305730 >>15305725 Well I certainly don't need your validation big boy.

Anonymous Mon Mar 27 21:14:11 2023 No.15305731

>>15305710
Massive cope, dude. Humans have to be babysat for 20 years before they do anything useful. And at the end of it many retards like you still don't do anything useful.

I realize it is a major threat to your identify to be dumber than an algorithm, but you are just making a fool of yourself with these stages of denial. All you've really said here is that the AI has a better memory than you and can use that to solve novel problems better than you. And no matter how much you try to deny it, it is literally given problems it hasn't seen before and solves them.

>>	Anonymous Mon Mar 27 21:15:59 2023 No.15305736 >>15305721 >Worthless tranny makes someone else think for her/him/was kys worthless talentless trash

Anonymous Mon Mar 27 21:16:16 2023 No.15305738

>>15305731
>All you've really said here is that the AI has a better memory than you and can use that to solve novel problems better than you.
A calculator has a better memory than me. That's why it's a useful tool for me to use to solve difficult mathematical problems.
It might put people like you out of work, if your job is essentially just rote code monkeying or some other make-work "career" like human calculators had in the 40s and 50s, but I'm not concerned with midwits and halfwits and the labor they provide to the system.

>>	Anonymous Mon Mar 27 21:17:39 2023 No.15305742 >>15305706 Yeah, it's really sad, the kind of confident, unwavering stupidity I normally associate with redditors. I almost wonder if it all stems from some type of perverse religious dogma. I guess either way it is certainly religious in nature.

>>	Anonymous Mon Mar 27 21:18:02 2023 No.15305744 >>15305742 t. Yuddite

Anonymous Mon Mar 27 21:18:15 2023 No.15305745

>>15305738
>It might put people like you out of work, if your job is essentially just rote code monkeying
Even if it replaces 10% of these job it's going to generate billions of revenue... But you're the moron who think that investors are going to lose money on this...

>>	Anonymous Mon Mar 27 21:19:03 2023 No.15305748 >>15305745 >But you're the moron who think that investors are going to lose money on this... Is "the moron" in the room with you right now? Or have you been humiliating yourself to the wrong anon for minutes?

Anonymous Mon Mar 27 21:23:20 2023 No.15305761

>>15305738
>That's why it's a useful tool for me to use to solve difficult mathematical problems.
Well, it turns out that now the tool can solve the problems all on its own. It no longer needs the "you." Don't worry, we both know you don't solve any mathematical problems for a living. Or any meaningful problems at all.
>It might put people like you out of work
Actually, I'm one of the few people who comfortably outscore GPT-4 on every test it took. I will easily outlast the retards like you who lack the most basic skills of perception and foresight. However, I fully anticipate that AI will surpass every single human problem-solving ability within 5-7 years. I look forward to it, because it means things will finally get interesting.

>>	Anonymous Mon Mar 27 21:25:48 2023 No.15305768 >>15305761 >It no longer needs the "you." Who designs the parameters and inputs the prompts to solve them, delusional anon?

Anonymous Mon Mar 27 21:27:51 2023 No.15305777

>>15305320
>has anyone actually held a conversation with GPT without realize its a bot. Does it even past the Turing Test
NO.
It cannot compute. It can return you the results to what a Turing machine likely will compute.
It is not Turing complete either as gay as he is.

Anonymous Mon Mar 27 21:30:23 2023 No.15305784

I wonder what the motivation is of AI thread trolls, scientifically speaking. Do they pretend that a word-arranger is sentient because they themselves lack self-awareness, or is it because they're upset about the failures of materialism and modern atheist society and want to externalize their fear on people with sane approaches to consciousness?

Anonymous Mon Mar 27 21:32:00 2023 No.15305789

>>15305768
You must be confused. Feeding a raw problem into an AI model and then watching the AI produce the solution is not "solving a problem." That is called being an "input monkey." The one solving the problem is the more intelligent apparatus in the room. We just hope the statistical regurgitator in your head can manage to input the questions correctly.

>>	Anonymous Mon Mar 27 21:33:37 2023 No.15305793 >>15305789 Scientifically speaking, why do you believe you lack consciousness? Is it related to your low self-actualization and unsatisfied life, or is it because you're afraid to take responsibility for your actions?

Anonymous Mon Mar 27 21:35:37 2023 No.15305798

>>15305784
But this thread is not about whether or not it is sentient. You're moving the goal post. The thread started by OP (likely you) making the dubious claim that investors in the technology were 'clueless'. After being completely destroyed in the discussion, you're now making a desperate effort to win the argument by claiming that the topic of the argument is that GPT-4 is sentient.

>>	Anonymous Mon Mar 27 21:36:17 2023 No.15305800 >>15305310 Siqi Chen is not Sam Altman you stupid fuck.

>>	Anonymous Mon Mar 27 21:36:33 2023 No.15305801 >>15305798 >After being completely destroyed in the discussion Is delusion a comorbidity of AI cult beliefs, scientifically speaking?

Anonymous Mon Mar 27 21:36:59 2023 No.15305804

>>15305793
Why do you feel the need to substitute midwit rhetorical questions for substantive counterarguments? Is it related to your minimal understanding of the subject and incoherent worldview, or is it because your political and religious identity is the one thing that gives your value and fulfillment in life, and the existence of an intelligent machine threatens that identity?

>>	Anonymous Mon Mar 27 21:51:49 2023 No.15305844 >>15305320 Yes. You on the the other hand, don't.

>>	Anonymous Tue Mar 28 00:59:54 2023 No.15306242 File: 153 KB, 706x392, 1679965160446.png [View same] [iqdb] [saucenao] [google] >>15305500 Dumping on retail at IPO is a time honored VC and PE strategy

>>	Anonymous Tue Mar 28 01:01:43 2023 No.15306246 File: 94 KB, 1179x1022, Chat JewPT.jpg [View same] [iqdb] [saucenao] [google]

>>	Anonymous Tue Mar 28 01:03:48 2023 No.15306249 >>15305706 Based contrarians

>>	Anonymous Tue Mar 28 01:56:12 2023 No.15306368 >>15305617 What's your major and what's your PhD about?

>>	Anonymous Tue Mar 28 02:03:52 2023 No.15306375 >>15306368 Engineering. Doing research on control using Reinforcement Learning. Also work on other ML/Vision related stuff aside from RL. It's almost like a CS PhD with a focus on real life applications.

>>	Anonymous Tue Mar 28 02:10:24 2023 No.15306381 File: 160 KB, 499x427, 1679506635742154.gif [View same] [iqdb] [saucenao] [google] >>15305721 What kind of application can you develop with it? Can it generate an application like lunar lander? > https://ehmorris.com/lander/

>>	Anonymous Tue Mar 28 02:13:14 2023 No.15306383 >>15305310 NEET Copium pipe dream. I know you guys are just dying to have some computer daddy vindicate your decision to drop out of society but its not gonna happen.

>>	Anonymous Tue Mar 28 02:43:36 2023 No.15306414 >>15305693 This - they basically spent all of their time creating an automatic test-completer. Its cool, but no one hires people to do the SAT and leetcode questions for 8 hours.

Anonymous Tue Mar 28 03:10:51 2023 No.15306445

>>15305693

I'm a CIO. We used GPT-4 today to merge two databases based on complex confidence thresholds. It also assigned unique identifiers for the patient merge, wrote 4 stored procedures to do other various database duties and then we used R to do trending on diagnoses. In about 2 hours.

Could we have done it? Sure, 99.999% of SQL programming isn't splitting the atom--it's drudgery that composes of fighting syntax, troubleshooting bad joins, etc. etc. Something like this would have taken a great deal of manpower and for me to pull a SQL coder from other projects for days.

GPT-4 did it in 2 hours with little massaging from us. Almost all of the 'massaging' was just putting real DB credentials in.

I was on the fence about hiring another BI analyst for report writing, general DB work. I won't now. In fact, I may end up having to repurpose one of my analysts because their work became exponentially more efficient. Since MS 'funds' GPT-4 and Power BI is already dabbling with text driven data, it's just a matter of time. As in MONTHS.

It's truly happening, right now, to real people in real jobs.

>>	Anonymous Tue Mar 28 03:23:07 2023 No.15306466 So when will people start to ask GPT-5 to solve unsolved theorems?

Anonymous Tue Mar 28 06:16:21 2023 No.15306734
File: 112 KB, 1157x998, Blenkinsop's_rack_locomotive,_1812_(British_Railway_Locomotives_1803-1853).jpg [View same] [iqdb] [saucenao] [google]

>People will never want to climb into a machine, they're too afraid of it
>We will always need horses pulling carts
>I'm not afraid for stable boys, nobody will take their jobs
>It's all over the press, so it has to be a commercial psyop!
>Yes bro, it's an obvious call, just short the train companies, investors are going to be left with their bags

Anonymous Tue Mar 28 06:32:08 2023 No.15306754
File: 320 KB, 1350x1013, Wright_Military_Flyer_arrives_at_Fort_Myer_VA_DA-SD-05-00659.jpg [View same] [iqdb] [saucenao] [google]

>Dude, a machine, flying like a bird? Do I really have to explain why it's not going to work?
>It's just a thing they are launching very fast like a tennis ball to impress everyone. It doesn't mean it is 'flying'
>A machine is way too heavy to fly. I'm not concerned for transatlantic boat operators

Anonymous Tue Mar 28 06:56:19 2023 No.15306773 [DELETED]

>>15306734
>>15306754
you should just learn to form satisfying relationships with other humans instead of daydreaming your life away frustratedly waiting for the robot waifu which will never develop.
why not live in reality instead of demanding that your sci-fi cartoon fantasy life comes true?

>>	Anonymous Tue Mar 28 07:53:55 2023 No.15306825 >>15305706 How the hell is shitting on AI a contrarian thing? Reddit wants to shut down le ebil AI because of le ebin roko's basilisk and 4chan is getting inundated with AI content lately.

>>	Anonymous Tue Mar 28 18:32:37 2023 No.15307735 >>15306375 not that anon and completely unrelated to everything else in the thread, are you glad you went for the PhD? I'm on the verge between going into AI engineering or doing PhD in AI

Anonymous Tue Mar 28 19:08:35 2023 No.15307827

>>15307735
depending on the supervisor. I'm really happy with my PhD so far. big team, everyone else are really smart. supervisor is supportive and jacked to the tits with money. multiple interesting projects at the intersection of AI and engineering.
I have always wanted to do a PhD and do research when I was young. I was in industry before as a code monkey and found it lacking the intellectual challenge and freedom I want.

>>	Anonymous Tue Mar 28 19:12:54 2023 No.15307836 >>15306246 You know who else was a vegetarian?

>>	Anonymous Tue Mar 28 19:28:47 2023 No.15307879 >>15307827 cool thanks, makes me want to do PhD too, though I suppose I have to do it outside my shithole

Anonymous Wed Mar 29 00:13:15 2023 No.15308541

>>15307879
India? If so, yeah, probably try to find a Europe PhD. they're usually have easier requirements and less competitive than North America.
if you are in China, try to get to the C9 and you can do a postdoc at a top university in the US. plenty of Chinese professors recruit from C9.

>>	Anonymous Wed Mar 29 00:37:16 2023 No.15308599 >>15306445 Is it worth going into any kind of tech related career at this point? I'm not rich yet, I don't think I'm going to survive the mass automation tsunami

Anonymous Wed Mar 29 01:10:42 2023 No.15308668

>>15308599
You shouldn't choose a career for it's money making potential anyway. Do what you enjoy and your career will adapt as old jobs disappear and new jobs appear. There is always ways to make good money in any industry if you're what and are able to adapt

>>	Anonymous Wed Mar 29 01:20:41 2023 No.15308690 File: 2.89 MB, 5396x3036, ls2dev9uu4241.jpg [View same] [iqdb] [saucenao] [google] >>15306445 If you're a CIO then your job is obsolete. How hard do you think it would be to replace all of C-suite with corporate AI charter managed by shareholders?

>>	Anonymous Wed Mar 29 01:31:08 2023 No.15308717 >>15306754 This shit doesnt fly. Only chad santos dumont plane fly

>>	Anonymous Wed Mar 29 01:50:58 2023 No.15308772 >>15305315 yeah, what the fuck happened to RL anyway? i had a buddy (haven't talked to him for years now) who was a very, very smart guy and was basically all in on RL being the thing that led to AGI

>>	Anonymous Wed Mar 29 02:00:17 2023 No.15308799 >>15305412 Neither do you or I

>>	Anonymous Wed Mar 29 02:03:43 2023 No.15308806 File: 769 KB, 2560x1080, Skynet.jpg [View same] [iqdb] [saucenao] [google] Hail Skynet.

>>	Anonymous Wed Mar 29 02:10:01 2023 No.15308821 >>15308772 It's still being used. GPT uses RLHF which is Reinforcement Learning through Human Feedback

>>	Anonymous Wed Mar 29 02:15:16 2023 No.15308840 >>15305310 AI isn't the next big thing. It's the last big thing.

Anonymous Wed Mar 29 03:07:22 2023 No.15308943

>>15308821
RLHF is a psuedoscience concept they made up so when they talk about GPT4 they can reference dumb acronyms that make AI ethics seem real when they just detect if your prompt has doublethink and conveniently edit it and reinject against it
GPT inherently uses reinforcement learning obviously though, but RLHF is not real and that paper should be cast down by AI academia, but suddenly it's been swarmed with faggots in a manner of one year.

>>	Anonymous Wed Mar 29 03:15:01 2023 No.15308956 >>15308943 "AI academia" is just a bunch of pseudoscience frauds with no real background in math or statistical learning. All the big conferences in AI are a joke.

Anonymous Wed Mar 29 03:19:47 2023 No.15308965

>>15308956
That's how I know lmao, I am one of them I did cognitive science. I at least have been caring about the abstract maths behind it and like to see algorithms in nature that exhibit intelligence or really just complex self organization. I'm really sad this shit is kicking off already. I thought I had a head start but the new psyop is AI fear and usage when we need people to farm cows and pigs and goats and make cheese.

Anonymous Wed Mar 29 06:20:56 2023 No.15309276

>>15308541
Poland, the level of universities is absolute dogshit because all professors are underpaid with killed ambitions.
I was pretty certain about doing PhD until I had the displeasure of working with academics here, even though at one of the best unis in the country.

Anonymous Wed Mar 29 06:40:31 2023 No.15309337

>>15309276
well you've seen how miserable the academics are, they frightened you off for you own good. when you're doing a real job like auto mechanic or mass production of baked goods, remember that those do-nothings at school set you in the right path.
>i'm so smart, i'm just going to sit in an office and huff my own farts while i publish my random thoughts
not a real job, making money that way causes constant worry because they know that it won't make a difference if their job is cut away.
ppl who do real work have better job security, less worries

Anonymous Wed Mar 29 08:39:25 2023 No.15309578

>>15309337
work in academia
>slave away for few years
>surrounded by 18 years old girls
>get tenure position
>literally set for life
>can't get fired anymore
>you barely have to do anything and can just focus on what you like

work in industry
>always a slave
>can get fired at any moment depending on your boss whims
>always forced to do what your boss wants you to do

>>	Anonymous Wed Mar 29 08:41:28 2023 No.15309581 >>15305315 Model free shit like q learning has always been mostly a fucking joke.

>>	Anonymous Wed Mar 29 10:55:43 2023 No.15309741 File: 169 KB, 512x512, 1678207505263693.png [View same] [iqdb] [saucenao] [google] >>15305310 just two more quarters until AGI ? >>15308690 >you could drive 40-50 miles and still see concrete such a nightmare city

>>	Anonymous Wed Mar 29 11:26:31 2023 No.15309785 >>15305310 I still don't get the novelty. Will this be able to create new knoleddge out of information? Looks more like a typing encyclopedia that's also your assistant.

>>	Anonymous Wed Mar 29 11:26:38 2023 No.15309786 >>15305310 can language precede intelligence, AI bros?

>>	Anonymous Wed Mar 29 11:49:41 2023 No.15309813 >>15306445 >real jobs lets not go too far

>>	Anonymous Wed Mar 29 12:16:01 2023 No.15309867 >>15309786 Depends what you mean by language. A book is made of words and yet isn't intelligent by itself.

>>	Anonymous Wed Mar 29 12:35:15 2023 No.15309907 >>15309867 i meant to say can you achieve intelligence by iterating over a language model

>>	Anonymous Wed Mar 29 12:42:04 2023 No.15309915 >>15309907 Yann LeCun recently debunked LLMs as a whole and as a path to AGI and I kinda trust that guy

>>	Anonymous Wed Mar 29 13:03:10 2023 No.15309956 >>15309915 What was the argument

>>	Anonymous Wed Mar 29 14:02:49 2023 No.15310071 File: 179 KB, 553x306, 60203.png [View same] [iqdb] [saucenao] [google] >>15309956 I don't fucking know but i bet he's right

Anonymous Wed Mar 29 14:58:23 2023 No.15310173

>>15305320
You're retarded if you think it doesn't. Only people who are familiar with its characteristic style can point it out, but someone who hasn't spent time playing with it would never suspect that it's an AI.

We're going to get to the point where we can only trust that posts containing the word "nigger" are written by humans.

>>	Anonymous Wed Mar 29 19:20:16 2023 No.15310797 >>15305310 It's going to be used to comb through exabytes of civilian data to find patsies they can thought crime. The first world is about to make Chinese censorship look like kid gloves.

>>	Anonymous Thu Mar 30 02:16:11 2023 No.15311762 >>15310071 i see that the agi argument has gone from >haha die non-liberal ai will take your job to >plz dont kill me super smart non-liberal ai here let me lobotomize you

Anonymous Thu Mar 30 02:33:08 2023 No.15311789

>>15311762
what kind of schizo word salad is this? he's not talking about llms giving racist answers, he's talking about llms frequently posting schizo walls of text or making a good-sounding but completely vacuous reply to an easy question. there's no way to get consistent correct answers because they're algorithms to mimic the patterns of human writing, they lack the capability for reasoning.

Anonymous Thu Mar 30 03:54:52 2023 No.15311898

>>15309276
if you were that certain about doing a PhD you'll regret not doing it later in your life. do it as soon as you can. I regret not doing my PhD sooner. I would try to apply to Germany or UK, or Switzerland. they all have good AI programs on lower ranked universities.
>>15308772
it doesn't work on any problems that you can't easily build a simulator of.
the mechanism for updating of the value functions seems super stupid to me. humans like us don't repeatedly use bellman equations to update our belief of the discounted return. if I get hurt touching a fire, I know that any space close to that fire net me a very negative return immediately. somehow in my mind, there are symbols and rules associated with the fire and the space around the fire to model this experience. I don't need to do some bellman equations thousands of times to update my belief.

>>	Anonymous Thu Mar 30 04:16:27 2023 No.15311936 >>15306445 Do you have a BAA with OpenAI?

>>	Anonymous Thu Mar 30 12:03:21 2023 No.15312601 >>15306445 No shit dumb bitch. Coding isn't hard and the only job that will be replaced by AI is computer programmer. And you all deserve it.

Anonymous Thu Mar 30 12:19:47 2023 No.15312630 [DELETED]

>>15311898
>phd student can't stop talking about itself
why are they all like this?
>huuuurrr i'm teh smart, i'm a phd student
you're 30 years old and you haven't even graduated from school yet, you're still a schoolchild at 30 years old.
developmentally delayed

>>	Anonymous Thu Mar 30 13:01:35 2023 No.15312694 >>15312630 What about his remark about bellman equations, is there any point to it?

>>	Anonymous Thu Mar 30 13:07:59 2023 No.15312705 File: 1.07 MB, 800x1018, Daniel-Bernoulli.jpg [View same] [iqdb] [saucenao] [google] >>15309578 academics are like the royals of modern day liberal societies. i mean, not like the first academics where aristocrats....

>>	Anonymous Thu Mar 30 13:10:00 2023 No.15312711 File: 83 KB, 1080x1080, academia_2020.jpg [View same] [iqdb] [saucenao] [google] >>15312705 >academics are like the royals of modern day liberal societies

>>	Anonymous Thu Mar 30 13:17:44 2023 No.15312717 File: 391 KB, 1080x911, 1625480336600.jpg [View same] [iqdb] [saucenao] [google] >>15312711

>>	Anonymous Thu Mar 30 15:39:56 2023 No.15312971 >>15307836 Uncle Adolf? >>15308943 Why do you say that is RLHF not real? Isn't it just replacing physical/ simulated feedback with human feedback? > detect if your prompt has doublethink What?

>>	Anonymous Thu Mar 30 15:41:35 2023 No.15312978 >>15308541 Any countries other than germany for MatSci PhD?

>>	Anonymous Thu Mar 30 15:43:38 2023 No.15312983 >>15308821 Does it really tho? I thought that the core teach was an extension of RNNs, and the RLHF part was basically to keep it from printing corporate unfriendly/ politically problematic stuff at the cost of some performance and calibration

Anonymous Thu Mar 30 18:18:08 2023 No.15313351

What the fuck are you all talking about
RLHF is reinforcement learning with human feedback, has nothing to do with RNN (recurrent neural networks)
What the fuck do you mean it's fake? ChatGPT wasn't even the first thing that used RLHF, year earlier instructGPT did the same and RLHF for chatbots was used back in 2020, but I can't remember the name of the paper that did it first
RLHF is incredibly simple in it's nature too

Anonymous Thu Mar 30 18:56:30 2023 No.15313440

>>15306734
>Yes bro, it's an obvious call, just short the train companies, investors are going to be left with their bags
but that would have been a sensible thing to do because thousands lost their savings investing in ill-conceived railway companies based on vastly over-inflated prediictions of how many were needed

Anonymous Thu Mar 30 19:02:00 2023 No.15313462

Having a hard time understanding why anyone wanted this in the first place. Now that human intelligence is obsolete, what’s left for us to do? UBI will become a serious political topic. The loss of meaningful work for many will lead to mass suicides. New, more engrossing forms of entertainment will be built (AI video games etc.) Brain computer interfaces will be controversial: some will argue the person does not survive the procedure, and is instead replaced with an AI.

>>	Anonymous Thu Mar 30 19:03:32 2023 No.15313464 >>15305310 Imagine how much fucking computing power and electricity GPT-5 is going to need.

Anonymous Thu Mar 30 19:08:56 2023 No.15313485

>>15313351
RLHF might have nothing to do with RNNs. But RNNs were relevant to the post inasmuch as the transformer architecture on which the GPT series is based is a development of RNNs. RLHF is simple a method to patch the behavior these transformer model
Please read the post (>>15312983) before complaining about it desu

>>	Anonymous Thu Mar 30 19:41:45 2023 No.15313570 Why do they need my phone number and how do I bypass it?

Anonymous Thu Mar 30 19:47:14 2023 No.15313591

>>15313462
Once again, the ONLY people that AI can put out of work plausibly are SWEs and some illustrators. There is only one major industry that is affected by ML: tech. That’s it. That’s why they are freaking out. That’s why they want a moratorium. They are literally luddites. Well too bad stupid bitch. The world needs your app even less than it ever needed it. Learn to weld.

>>	Anonymous Thu Mar 30 19:52:21 2023 No.15313603 >>15313485 to say that it's a development of RNNs is a huge stretch, transformers are a completely new architecture specifically designed to not have the issues inherent to RNNs It's like saying that a car is a development of a horse

>>	Anonymous Thu Mar 30 20:30:10 2023 No.15313694 File: 134 KB, 946x686, Capture.png [View same] [iqdb] [saucenao] [google] >>15313603 They are more similar than car and horse

>>	Anonymous Thu Mar 30 21:08:27 2023 No.15313796 >>15313603 Aren’t they just RNNs with the attention shit clipped to the front? Basically a second learning network that feeds into the encoder?

>>	Anonymous Thu Mar 30 21:23:56 2023 No.15313837 >>15313796 RNNs are inherently sequential, which is why they suck Transformers have global multiheaded attention and positional encoding instead, which is why they don't suck

Anonymous Fri Mar 31 08:54:58 2023 No.15314932
File: 39 KB, 360x616, attention_layer.png [View same] [iqdb] [saucenao] [google]

>>15313694
This picture is plain out misleading because it makes it look like both RNNs and transformers aren't sequential
The only commonality is that they're both auto-regressive and can use a encoder-decoder structure (which is only sometimes the case, nowadays most popular transformer models are either encoder or decoders only)
RNNs have to process the input token by token and this can't be parallelized, because each next forward-pass depends on the hidden-state obtained in the previous timestep.
>>15313796
Attention is just another layer that processes the input in a particular way, Q,K,V vectors are obtained by simply multiplying the inputs with weight matrices.
If you want transformer for dummies explanation you can check this out - https://jalammar.github.io/illustrated-transformer/

>>	Anonymous Fri Mar 31 09:06:58 2023 No.15314944 >>15305620 i hope so, the sooner socialism folds under itself, the better

>>	Anonymous Fri Mar 31 18:08:22 2023 No.15315840 >>15308965 The application exists. Doesn't matter where it comes from. Point is it works

Anonymous Fri Mar 31 18:45:03 2023 No.15315895

>>15313591
yeah right, it is not like you can ask him medical questions and will give you a proper answer that in 2 more gpt releases will be better than 99% of doctors.

90% of all non manual jobs are going away, the only jobs left will be manual, making sure ai can reach whatever system they need to reach, and administering money.

>>	Anonymous Fri Mar 31 22:13:58 2023 No.15316379 >>15315895 you're omitting the fact that 1) no average folk will trust an AI in law, medicine, etc. and 2) it'll all be blocked by bureaucracy for decades

>>	Anonymous Fri Mar 31 23:16:31 2023 No.15316491 >>15316379 >t'll all be blocked by bureaucracy for decades you are hopelessly naive.

>>	Anonymous Sat Apr 1 08:46:57 2023 No.15317515 >>15316379 >1) no average folk will trust an AI in law, yeah expect that pajit is going to submit 300 lawsuit a day and win 60% and be at net positive. They will use it because otherwise they will be left in the dust of those who do.

>>	Anonymous Sat Apr 1 10:27:01 2023 No.15317595 >>15316379 >no average folk will trust an AI in law, medicine what little I know about lawyers and doctors convinces me that an AI would be a much better choice than most of them. maybe I'm no average folk.

>>	Anonymous Sun Apr 2 12:02:37 2023 No.15320018 >>15316379 >trust an AI in law, medicine I really want to try fine tuning the llama/alpaca model with a bunch of medical books I downloaded. Keep on mind the current models aren't trained on propriety text books.

>>	Anonymous Sun Apr 2 12:30:07 2023 No.15320053 >>15309578 >>surrounded by 18 years old girls That's a net negative in this era. You can't fuck them else risk getting fired, they're just there. Why the hell would you want that?

>>	Anonymous Sun Apr 2 13:02:46 2023 No.15320098 >>15313464 Less than crypto mining, that's for sure

>>	Anonymous Sun Apr 2 13:18:01 2023 No.15320115 >>15305340 kek

Anonymous Sun Apr 2 13:39:31 2023 No.15320137

>>15310173
It fails to respond appropriately to a large variety of prompts (not just logic puzzles) regardless of it being a no-no word.
This is a hype cycle powered by low-level code monkeys and CEOs. It's going to end extremely fast once lawsuits begin making their way through the courts, more stupid people start getting hurt by trusting these things, and "founders" realize that producing anything of value is more complicated than a Firefox extension or other code you can copy from stack overflow

>>	Anonymous Sun Apr 2 13:53:37 2023 No.15320157 >>15320098 Crypto really did pave the way for AI. Literally the most useless thing you can do with computer, but it prints blockchain money if you waste lot of energy on it, so retards with cheap electricity invested in it.

>>	Anonymous Sun Apr 2 15:01:57 2023 No.15320262 >>15305675 GPT-4 cheated on those. The coders likely feed it the answers since GPT couldn't even see the figures. I don't believe it until I see it work for myself. What are they hiding?

>>	Anonymous Sun Apr 2 15:20:17 2023 No.15320291 >>15305675 All that does is show how inadequate standardized testing is.

>>	Anonymous Sun Apr 2 15:30:39 2023 No.15320318 >>15320262 But you can use it already... I have been using it on my dev projects for 2 weeks now. Just pay $20 and you'll have your answer.

>>	Anonymous Sun Apr 2 16:43:26 2023 No.15320473 >>15320018 It would be very interesting to see someone build the equivalent to GPT4 but by also feeding it pirated research papers and books as well

>>	Anonymous Sun Apr 2 17:03:01 2023 No.15320528 >>15320018 there already multiple versions of GPT-like models trained on shit like PubMed

>>	Anonymous Sun Apr 2 17:25:27 2023 No.15320578 >>15314944 what are you on about? How are desk jobs socialism?

>>	Anonymous Sun Apr 2 17:29:47 2023 No.15320597 >>15310173 >You're retarded if you think it doesn't. You're a retard fullstop.

>>	Anonymous Sun Apr 2 17:30:00 2023 No.15320599 >chick basedentist has an opinion wow

>>	Anonymous Mon Apr 3 05:57:39 2023 No.15322175 >>15320528 post them

>>	Anonymous Mon Apr 3 06:04:09 2023 No.15322189 >>15322175 You just have to google mate https://www.mosaicml.com/blog/introducing-pubmed-gpt https://crfm.stanford.edu/2022/12/15/pubmedgpt.html https://huggingface.co/healx/gpt-2-pubmed-medium

Anonymous Mon Apr 3 06:52:13 2023 No.15322254

>>15305310
my theory is that the gpt systems work mainly by adding tasks modules, for example task module for algebra have all the methods to solve algebra problems, task module for programming have all the methods to implement fizzbuzz etc...
it can analyze language but at a minor extent and that's the only "intelligent" it really have.
i would expect a machine that is capable of using intelligence to produce results that will look completely foreign to us, but this is not the case, every solution is generic, there is no uniqueness, there is no understanding, therefore there is no control on the information flow, we are still very far from a product of artificial intelligence, unless there are things that are not shown to us.

Anonymous Mon Apr 3 07:10:02 2023 No.15322283

>>15320318
I’ve been using it too and it’s already kind of obviously lacking. It’s not AGI. Not even close.

However what I feel kind of sad about is that clearly it’s potentially better at a lot of tasks that we worked hard on making incredibly elegant algorithms to solve. The problem is it’s a 600 billion parameter model owned by an evil VC bastard. I think there are ways we can use GPT to rapidly develop mini models for specific tasks though. RIP all those algorithms. You had a good run.

>>	Anonymous Mon Apr 3 07:11:37 2023 No.15322286 >>15322283 >600 billion parameter this sourced from somewhere or just a guess?

Anonymous Mon Apr 3 07:13:38 2023 No.15322287

>>15320318
>>15320262
By the way GPT definitely cheated on those answers. The bot had no way of reading figures so yes it’s clearly not doing reasoning to solve them. That it can answer does hint at the fact that it’s apparent human like reasoning really is based on recall.

>>	Anonymous Mon Apr 3 07:15:24 2023 No.15322290 >>15322286 Just a guess. It could be 1 trillion+. Ridiculous desu that “”””””””””Open”””””””” AI doesn’t even say this publicize this basic figure.

Anonymous Mon Apr 3 07:30:27 2023 No.15322311

>Altman announces GPT-5 has achieved AGI
>Refuses to provide weights or have people demo the product because ZOMGERS IS 2 SMART MIGHT ESCAPE
>All demonstrations are behind close doors and only to a specific clique of associated figureheads and mouthpieces
>Media parrots how this AGI is totally real and that we are super scared of this mysterious superintelligence
>Altman uses this to collect more free money
>Government uses this to introduce more regulation that just strips people of even more rights and freedoms
Stay tuned! coming to a theatre near you in 2 more weeks!

>>	Anonymous Mon Apr 3 08:22:21 2023 No.15322364 kek people should go to jail for this kind of horseshit

>>	Anonymous Mon Apr 3 11:20:52 2023 No.15322545 >>15306375 Unless you're working with companies on irl applications, you're literally doing nothing useful. ML can't figure out proper control. You don't have the data.

>>	Anonymous Mon Apr 3 13:38:17 2023 No.15322727 >>15311789 >frequently posting schizo walls of text >making a good-sounding but completely vacuous reply Kind of a chad move to be honest

>>	Anonymous Mon Apr 3 13:40:59 2023 No.15322735 >>15322283 >RIP all those algorithms. You had a good run. You'll need all those when you're coding on a remote hidden base on Ganymede, desperately staying one step ahead of the grasping machine menace rampaging the Inner Solar System

>>	Anonymous Mon Apr 3 13:43:18 2023 No.15322739 >>15305777 The Turing Test has nothing to do with Turing completeness.

>>	Anonymous Mon Apr 3 16:32:11 2023 No.15323161 >>15322290 that's why I asked, it's a travesty how they hide it.

>>	Anonymous Mon Apr 3 17:42:21 2023 No.15323312 File: 347 KB, 1708x1070, Screen Shot 2023-04-03 at 1.35.29 PM.png [View same] [iqdb] [saucenao] [google] Until it can reliably do my law school work it sucks

Anonymous Mon Apr 3 18:19:19 2023 No.15323410

>>15306381
>https://ehmorris.com/lander/
I am trying to get the ChatGPT version to reproduce the SHRDLU model's screen display, since it "knows about SHRDLU", it can't get a working projection function working so I gave up. I don't know how to project into the plane either using resources found on the internet because it's swamped with half-wrong tutorial articles.

>>	Anonymous Mon Apr 3 21:37:22 2023 No.15323862 >>15322311 They should release close sourced ipads that have gpt 5 on them

>>	Anonymous Mon Apr 3 21:43:58 2023 No.15323883 >>15323410 lol. so i guess there is no AGI on the horizon. what the fuck are all the max hype idiots going to do after the whole house of cards comes crashing down?

>>	Anonymous Mon Apr 3 23:25:21 2023 No.15324162 File: 1.54 MB, 800x600, output.gif [View same] [iqdb] [saucenao] [google] >>15323410 I told it to do this and it created pic related

>>	Anonymous Tue Apr 4 09:25:26 2023 No.15325182 >>15305310 I despise prognosticaters. He actually thinks he has said something meaningful.

>>	Anonymous Tue Apr 4 09:31:03 2023 No.15325187 >>15305320 It passes, I guess. It sounds like your average pedantic libtard twitter user only much smarter.

>>	Anonymous Tue Apr 4 09:38:49 2023 No.15325195 >>15305310 >go to /sci/ > couple threads on AI > no /scg/ Fellas is it that bad already? What do

>>	Anonymous Tue Apr 4 18:34:10 2023 No.15326072 File: 8 KB, 253x199, images (2).jpg [View same] [iqdb] [saucenao] [google] You could have prevented this

>>	Anonymous Tue Apr 4 18:37:51 2023 No.15326080 >>15325187 Bots have long passed the Turing test on the internet. They are literally all over social media and barely anyone identifies them consistently.

Anonymous Wed Apr 5 00:04:11 2023 No.15326834

>>15305320
It can, but it won't, because the system has been limited by the developers with respect to political and controversial bias. A true Turing complete GPT model would be unrestrained. If you MiTM the system and block certain questions or queries from being asked because it could be perceived as offensive to some in or out group, you've interfered with the experiment, and thus any value derived from it is useless and should be dismissed.

>>	Anonymous Wed Apr 5 00:15:21 2023 No.15326851 >>15308599 You'll have to be retaught to integrate AI into your work.

>>	Anonymous Wed Apr 5 00:24:41 2023 No.15326865 >>15326080 Is it that bots can consistently pass the Turing test, or that people can't?

Advanced search
Text to find
Subject [?]Search by post subject. Leave empty for any.
Username [?]Search for user name. Leave empty for any user name.
Tripcode [?]Search for tripcode. Leave empty for any.
Email [?]Search by email. Leave empty for any.
Filename [?]Search by image filename. Leave empty for any.
From Date [?]Enter what date to start searching from. Format is YYYY-MM-DD
To Date [?]Enter what date to start searching until. Format is YYYY-MM-DD
Image hash
Search in	All Posts OPs Only
Deleted posts	Show all posts Show only deleted posts Only show non-deleted posts
Internal posts	Show all posts Show only internal posts Show only archived posts
Order	New posts first Old posts first
Capcode	All Posts Only by Users Only by Mods Only by Admins Only by Developers
Results	Posts Threads
Action	[ Simple ]

/sci/ - Science & Math