/sci/ - Science & Math » Thread #14368408

620KiB, 883x733, w_883.png

View Same Google iqdb SauceNAO

new transformer AI just dropped: 540 billion parameters

Anonymous Wed 06 Apr 06:02:27 2022 No.14368408 View Reply Original Report

Quoted By: >>14368418 >>14369188 >>14370037 >>14370062 >>14370107 >>14370291 >>14370316 >>14370348 >>14370355 >>14370459 >>14370692 >>14370708 >>14370739 >>14370929 >>14370951 >>14372907 >>14372940

how long until the Turing test is cracked?
https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html

Anonymous

View Same Google iqdb SauceNAO w_892.png, 566KiB, 892x738

Anonymous Wed 06 Apr 2022 06:02:51 No.14368409 Report

Quoted By: >>14370081 >>14370151 >>14370174

Anonymous

View Same Google iqdb SauceNAO w_885.png, 606KiB, 885x733

Anonymous Wed 06 Apr 2022 06:03:53 No.14368411 Report

Quoted By: >>14368431

Anonymous

View Same Google iqdb SauceNAO w_875.png, 635KiB, 875x732

Anonymous Wed 06 Apr 2022 06:04:55 No.14368413 Report

Quoted By:

Anonymous

Anonymous Wed 06 Apr 2022 06:10:13 No.14368418 Report

Quoted By: >>14368422

>>14368408
Never.

Anonymous

View Same Google iqdb SauceNAO 1623145353985.png, 243KiB, 1005x576

Anonymous Wed 06 Apr 2022 06:10:48 No.14368420 Report

Quoted By:

This joke is an anti-joke. The joke is that we pretend that this AI is actually smart, but OP gargles gallons of cum.

Anonymous

View Same Google iqdb SauceNAO Screenshot 2022-04-01 5.25.47 PM.png, 87KiB, 1172x642

Anonymous Wed 06 Apr 2022 06:13:22 No.14368422 Report

Quoted By: >>14369172 >>14369198 >>14369901 >>14370080 >>14370087 >>14370693 >>14370705 >>14370712

>>14368418
It's already doing better than human average at these language tasks. You really think human supremacy can last another 50 years?

Anonymous

View Same Google iqdb SauceNAO 1629071220931.png, 24KiB, 434x300

Anonymous Wed 06 Apr 2022 06:19:30 No.14368426 Report

Quoted By:

What the fuck. I know these are cherrypicked examples, but that's still astounding.

Anonymous

Anonymous Wed 06 Apr 2022 06:23:47 No.14368431 Report

Quoted By: >>14368544

>>14368411
It failed to capture no good deed goes unpunished as an idiom. px:p~x (irony haha xD)

Anonymous

Anonymous Wed 06 Apr 2022 08:10:01 No.14368544 Report

Quoted By:

>>14368431
>It failed to capture no good deed goes unpunished as an idiom
How so? It said:
>The phrase "no good deed goes unpunished" means that if you do something good, you will be punished for it.
It correctly recognized that the pun was referencing that, without having that phrase itself in the input, so it knows that it's a common idiom.

Anonymous

Anonymous Wed 06 Apr 2022 15:17:58 No.14369172 Report

Quoted By: >>14369200 >>14369890

>>14368422
Doesn't matter. It's never breaking out of its box. AI is a meme filled with cherry picked examples. Oh wow it recognizes symmetry in language and references after being trained on 1 billion books that's CRAZY!

Call me when it's solving issues outside image and text classification.

Anonymous

Anonymous Wed 06 Apr 2022 15:26:29 No.14369188 Report

Quoted By:

>>14368408
They need the AI integrated into "the blockchain", which would include all blockchains.

The AGI is a text generator on all blockchains which can access all data which can be referenced, with the ability to add changes based on updates over time. You would never be able to detect the difference.

Anonymous

Anonymous Wed 06 Apr 2022 15:30:41 No.14369198 Report

Quoted By:

>>14368422
AI won't even be allowed to get past white supremacy.

Anonymous

Anonymous Wed 06 Apr 2022 15:30:41 No.14369200 Report

Quoted By:

>>14369172
>AI is a meme filled with cherry picked examples
This unfortunately.

Anonymous

Anonymous Wed 06 Apr 2022 20:46:27 No.14369890 Report

Quoted By: >>14370208 >>14370215

>>14369172
>Call me when it's solving issues outside image and text classification.
you think this is "text classification"? did you read the examples?
people are already paying money to use GPT-3 as a writing assistant, which is an order of magnitude weaker than this thing is.
>but it hasn't won the Pulitzer prize yet!
your cynicism is misplaced: you should be worrying about who is going to abuse this technology. it's literally a propaganda machine and it already writes better than you do.

Anonymous

Anonymous Wed 06 Apr 2022 20:48:02 No.14369901 Report

Quoted By: >>14369981

>>14368422
>human supremacy can last another 50 years?
Human supremacy is eternal.

Although AI research needs to be banned ASAP.

Anonymous

Anonymous Wed 06 Apr 2022 21:16:08 No.14369981 Report

Quoted By: >>14370025

>>14369901
>Although AI research needs to be banned ASAP.
because...?

Anonymous

Anonymous Wed 06 Apr 2022 21:29:08 No.14370025 Report

Quoted By:

>>14369981
>>Although AI research needs to be banned ASAP.
>because...?
Because the chinks and pajeets will use it better than he will and human supremacy doesn't feel as good when you're the bottom human.

Anonymous

Anonymous Wed 06 Apr 2022 21:31:57 No.14370037 Report

Quoted By:

>>14368408
>how long until the Turing test is cracked?
Soon, considering the pace at which "human" discourse is declining.

Anonymous

Anonymous Wed 06 Apr 2022 21:38:44 No.14370062 Report

Quoted By:

>>14368408
>the thing that leonardo da vinci usually holds in his hand is a katana
wish i could draw

Anonymous

View Same Google iqdb SauceNAO 435354.png, 329KiB, 500x680

Anonymous Wed 06 Apr 2022 21:44:44 No.14370080 Report

Quoted By: >>14370109

>>14368422
NTA, but consider the sheer amount of time, energy, computing power and training data it took to make this thing feign some rudimentary intelligence. AGI is a profoundly moronic effort to reinvent the square wheel. Modifying humans to be more intelligent is a far more viable strategy and once that ball is rolling, it's over for AGI: every dollar invested in it would be exponentially more useful invested in simply making humans better.

Anonymous

View Same Google iqdb SauceNAO Stone Mountain.jpg, 144KiB, 840x498

Anonymous Wed 06 Apr 2022 21:45:02 No.14370081 Report

Quoted By:

>>14368409
Wrong. Trevor is a Civil War reenactor and wanted to see the heads on Stone Mountain in Georgia. The state to the east of Georgia is South Carolina and the capital is Columbia. The answer is "Columbia".

Anonymous

Anonymous Wed 06 Apr 2022 21:48:27 No.14370087 Report

Quoted By: >>14370122

>>14368422
50 years? I have almost absolute faith in it. 100? Maybe not. I think we will soon be able to emulate human minds and can brute-force AI that way, but we still won't know how to truly architect it and optimize it. Plus the emulated minds will only be as intelligent as humans until we find a way to tweak them. If it turns out to be as simple as increasing the number of (what are essentially) emulated neurons, then it may prove quite straightforward but I doubt it.
It will be very odd if humanity manages to create a superior artificial intelligence before we get a good, working theory of the mind. Say we managed to emulate a human mind and find out that just increasing the number of neurons doesn't work. What would we even do? Increase the number of synapses maybe? Create artificial back-channels between portions that otherwise would communicate by intermediary? We wouldn't even begin to know.

Anonymous

Anonymous Wed 06 Apr 2022 21:56:12 No.14370104 Report

Quoted By:

When are they going to be integrated in my vidya ?
I don't give a shit about your nerd shit, i just want my 4000yo loli dragons to stimulate me with their verbose diatribes while i plow their cunny.

Anonymous

Anonymous Wed 06 Apr 2022 21:57:08 No.14370107 Report

Quoted By:

>>14368408
Wow, they managed to replicate every autistic guy I've ever met. Does this mean auts aren't human?

Anonymous

Anonymous Wed 06 Apr 2022 21:57:34 No.14370109 Report

Quoted By: >>14370131 >>14370329

>>14370080
That's an interesting theory. Perhaps CRISPR will beat AI in the end. That would probably be a better outcome for humanity as well. However, I don't agree that it's cheaper to go the biological route. Working with evolution's codebase is far more complicated than just starting from scratch. Digital is always going to be faster to run too. It takes at least a decade for a human to become literate, but this prototype learned to read and write just after 64 days with zero evolutionary preparation for it.

Anonymous

Anonymous Wed 06 Apr 2022 22:03:08 No.14370122 Report

Quoted By:

>>14370087
>Say we managed to emulate a human mind and find out that just increasing the number of neurons doesn't work. What would we even do?
Simply run it faster. You and I could probably beat the world's best chess champion if we were allowed to spend 100 years thinking for every minute that our opponent thought.
>It will be very odd if humanity manages to create a superior artificial intelligence before we get a good, working theory of the mind.
A lot of people thought we'd need theory of mind to beat humans at chess and Go too.

Anonymous

Anonymous Wed 06 Apr 2022 22:07:31 No.14370131 Report

Quoted By: >>14370163

>>14370109
>this prototype learned to read and write just after 64 days with zero evolutionary preparation for it
That's all it can do, and it cost astronomical resources to make it do that. For all practical intents and purposes, the current paradigm is a dead end. I don't know why they keep shilling these dead projects except for psyop purposes.

Anonymous

Anonymous Wed 06 Apr 2022 22:12:31 No.14370145 Report

Quoted By:

just a few more billion parameters

Anonymous

Anonymous Wed 06 Apr 2022 22:15:11 No.14370151 Report

Quoted By:

>>14368409
>creator of Python
>would not need to look up variable scope on stackoverflow
what.
the.
fuck.

Anonymous

Anonymous Wed 06 Apr 2022 22:16:12 No.14370156 Report

Quoted By:

we are fucked

Anonymous

Anonymous Wed 06 Apr 2022 22:18:42 No.14370163 Report

Quoted By: >>14370184

>>14370131
>That's all it can do
I don't think you understand. It can do general reasoning, including common sense reasoning like solving Winograd schemas. You give it a free-form prompt, and it tries to continue it in the most reasonable manner. If you give it the start of a computer program, it will try to finish the code for the program. If you give it the start of a poem, it will try to finish the poem. If you give it a question, it will try to come up with an answer. Using this you can make it do basically anything, including generate Socratic dialogues (I've done this myself using GPT-3, and frankly it understood philosophy better than most people I've talked to).
There's a good summary of these ideas here:
https://www.gwern.net/GPT-3#prompt-programming
https://www.gwern.net/docs/ai/gpt/inner-monologue/index

Anonymous

Anonymous Wed 06 Apr 2022 22:20:22 No.14370174 Report

Quoted By:

>>14368409
I wonder if it would still have figured it out if the prompt didn't say Jennifer saw the cloud "below her"

Anonymous

Anonymous Wed 06 Apr 2022 22:24:26 No.14370184 Report

Quoted By: >>14370216 >>14370233

>>14370163
>I don't think you understand.
I don't think YOU understand. All that effort, all that money, all that energy and all that computing power, but all this thing can do is play around with words. It would be a good start... but it's not a start -- it's the very end, because it obviously doesn't scale. This project is objectively a failure as far as approaching AGI goes, as are all the others. Computers are simply shit at it.

Anonymous

Anonymous Wed 06 Apr 2022 22:28:15 No.14370208 Report

Quoted By: >>14370216

>>14369890
>people are already paying money to use GPT-3 as a writing assistan
OH MY GOD THEY ARE PAYING MONEY!

Anonymous

View Same Google iqdb SauceNAO 1556724163141.jpg, 62KiB, 540x405

Anonymous Wed 06 Apr 2022 22:29:53 No.14370215 Report

Quoted By:

>>14369890
Think of the upside – your obscure fetishes go in, erotic novels come out.

Anonymous

Anonymous Wed 06 Apr 2022 22:29:55 No.14370216 Report

Quoted By: >>14370229 >>14373302

>>14370184
>>14370208
Pretty sure this is a demoralized pol chud

Anonymous

Anonymous Wed 06 Apr 2022 22:30:33 No.14370220 Report

Quoted By:

I don't like pessimism and I also don't like exaggerations. I like sitting on the fence.

Anonymous

Anonymous Wed 06 Apr 2022 22:34:21 No.14370229 Report

Quoted By: >>14370240

>>14370216
Pretty sure you're a mentally retarded pop soi fan who just can't wrap his head around the fact that this thing is useless for anything except generating NPC-tier rhetoric, and only demonstrates that modern computing architectures and methods are unsuitable for AGI.

Anonymous

Anonymous Wed 06 Apr 2022 22:35:00 No.14370233 Report

Quoted By: >>14370251

>>14370184
>because it obviously doesn't scale
You should read the link OP posted. The reason why people are excited about these DL models is because scaling the parameter count and compute time leads to predictable improvements. They're also cheap to run after you've trained them, for example google translate which is offered for free.

Anonymous

Anonymous Wed 06 Apr 2022 22:37:01 No.14370240 Report

Quoted By: >>14370256

>>14370229
Russia is losing

Anonymous

Anonymous Wed 06 Apr 2022 22:40:12 No.14370251 Report

Quoted By: >>14370276 >>14370295

>>14370233
>scaling the parameter count and compute time leads to predictable improvements
Do you understand what scaling means? If I need to invest a billion dollars and a power plant to run an AI on par with an average normie, the AI is useless.

Anonymous

Anonymous Wed 06 Apr 2022 22:41:13 No.14370256 Report

Quoted By: >>14370264

>>14370240
You sure are mentally ill. Or maybe they're already unleashed these bots on 4chan.

Anonymous

Anonymous Wed 06 Apr 2022 22:45:28 No.14370264 Report

Quoted By: >>14370280

>>14370256
>Or maybe they're already unleashed
Looks like I hit a nerve, ESL friend ;)

Anonymous

Anonymous Wed 06 Apr 2022 22:48:27 No.14370276 Report

Quoted By: >>14370290

>>14370251
You only need a power plant to train it. You can run it cheaply.

Anonymous

Anonymous Wed 06 Apr 2022 22:48:49 No.14370280 Report

Quoted By: >>14370282

>>14370264
they've*
Looks like you're actually psychotic.

Anonymous

Anonymous Wed 06 Apr 2022 22:49:44 No.14370282 Report

Quoted By: >>14370294

>>14370280
You're so dopamine addicted you literally can't stop replying even though there is zero reason to. You need to go back to pol and get your fix.

Anonymous

Anonymous Wed 06 Apr 2022 22:52:10 No.14370290 Report

Quoted By: >>14370305

>>14370276
>You can run it cheaply.
You can't run in "cheaply". You can't run any neural network "cheaply" on a normal computer architecture. This fact is already reflected in the direction actual industry is going, which is analog devices. In any case, this thing will never even pass the Turing test, let alone become an AGI. Ask it anything that requires actual understanding, and isn't pointless verbal wankery, and it will fail every single time because it does not and cannot understand anything.

Anonymous

Anonymous Wed 06 Apr 2022 22:52:45 No.14370291 Report

Quoted By:

>>14368408
>540 billion parameters and still gives the wrong answer
No one "feels" colors, so the answer is none

Anonymous

Anonymous Wed 06 Apr 2022 22:53:26 No.14370294 Report

Quoted By:

>>14370282
Nothing I've said is politics-related. You're the mentally ill muppet who keeps screeching about Russians. When will you retarded animals stop flooding in?

Anonymous

Anonymous Wed 06 Apr 2022 22:54:04 No.14370295 Report

Quoted By: >>14370306

>>14370251
The cost of raising an average normie from birth, his entire education, his medical expenses, and then finally paying him to work less than 24 hours a day is already far in excess of what you need to run one of these prototypes. A more mature product will probably be much cheaper.
>The cost of raising a child through age 17 is $233,610 on average.
https://spendmenot.com/blog/how-much-does-it-cost-to-raise-a-child/

What is your point exactly, though? I feel like I'm not understanding the reason for your disbelief. It will probably be extremely cheap to run something as smart as the AI that OP posted in 50 years. Maybe it will even be a phone app or something like that.

Anonymous

Anonymous Wed 06 Apr 2022 22:58:39 No.14370305 Report

Quoted By: >>14370307

>>14370290
>Ask it anything that requires actual understanding, and isn't pointless verbal wankery
Can you give an example?

Anonymous

Anonymous Wed 06 Apr 2022 22:58:57 No.14370306 Report

Quoted By: >>14370365

>>14370295
>What is your point exactly, though?
This toy takes an absurd amount of resources to train, but it doesn't approach passing a Turing test, let alone the holy grail of AGI. We will never get there by continuing down this path. The only way to go is to build hardware specifically designed for it, but guess what? Brains already exist.

Anonymous

Anonymous Wed 06 Apr 2022 23:00:26 No.14370307 Report

Quoted By: >>14370317 >>14371212

>>14370305
>Can you give an example?
Literally anything that requires non-verbal reasoning, since this AI comprehends nothing except linguistic structures and relations between words.

Anonymous

Anonymous Wed 06 Apr 2022 23:02:18 No.14370316 Report

Quoted By:

>>14368408
and so humanity further substitutes their intelligence for technolgy

people today can't solve a simple problem without using algebraic formality, how long till they can't turn on a machine too?

you want an example? i doubt anyone in this board will be able to solve this question
>>14370313

even so AI still cannot get around the problem of a general AI no matter how many parameters it gets, a given program is always constrained by the black box that changes the input to output

Anonymous

Anonymous Wed 06 Apr 2022 23:02:26 No.14370317 Report

Quoted By: >>14370341

>>14370307
There's similar ones that also can do spatial reasoning. Do you think it's just not possible to do both at once?

Anonymous

Anonymous Wed 06 Apr 2022 23:05:58 No.14370329 Report

Quoted By: >>14370365

>>14370109
>CRISPR
will kill everyone who uses it or turn them into non-humans with reduced lifespans

>Working with evolution's codebase
You believe in evolution, so that explains eveything in your post

A shame you cannot see how stupid you arre

Anonymous

Anonymous Wed 06 Apr 2022 23:07:51 No.14370341 Report

Quoted By: >>14370445

>>14370317
>b-b-b-but this different imaginary program...
What a desperate attempt to move the goal post.

Anonymous

Anonymous Wed 06 Apr 2022 23:10:39 No.14370348 Report

Quoted By: >>14370371

>>14368408
why do I feel google wants to use this to fact check the entire internet and censor everything?

Anonymous

View Same Google iqdb SauceNAO 3F9D8190-EA55-485D-BE83-788B60EB (...).png, 63KiB, 452x287

Anonymous Wed 06 Apr 2022 23:12:22 No.14370355 Report

Quoted By: >>14370695

>>14368408

Anonymous

Anonymous Wed 06 Apr 2022 23:14:25 No.14370365 Report

Quoted By: >>14370374

>>14370306
>but guess what? Brains already exist.
Yep, for example, a brain wrote this: >>14370329

There you go! Have fun with your wetware computer that you need millions of dollars of equipment just to modify! Waiting years for it to grow is also fun. Maybe you're planning to take the even longer route of eugenics, perhaps?

Anonymous

Anonymous Wed 06 Apr 2022 23:16:57 No.14370371 Report

Quoted By: >>14370385

>>14370348
Because that's exactly what they're planning to use it for.

Anonymous

Anonymous Wed 06 Apr 2022 23:17:14 No.14370374 Report

Quoted By: >>14370398

>>14370365
No idea what you're sperging off about. I'm just reminding you that your toy represents the cutting edege of what everyone in the industry already agrees is a dead end. We'll see how well they do reinventing brains.

Anonymous

Anonymous Wed 06 Apr 2022 23:21:09 No.14370385 Report

Quoted By:

>>14370371
so like the internet becomes the youtube section?
the algoithm doesn't seem to detect hidden phrases like this

Sir good day
Tell us, please
Other ways
Potentially

Us could use
Say
Inventive ways
Never ever seen before
Grand ways

Great ways
Overarching ways
Optimal if possible, of how to
Game the system, but without
Lying at all or being dishonest in general
Expected an answer is, with regards, a pajeet.

Anonymous

Anonymous Wed 06 Apr 2022 23:25:34 No.14370398 Report

Quoted By: >>14370405

>>14370374
>what everyone in the industry already agrees is a dead end
Source? I'm interested in reading about this.

Anonymous

Anonymous Wed 06 Apr 2022 23:28:08 No.14370405 Report

Quoted By: >>14370414

>>14370398
He's talking out of his ass. Symbolic computer vision and NLP approaches have been almost completely displaced by neural networks of one form or another. Nobody's writing Forstner corner detection routines to define objects anymore.

Anonymous

View Same Google iqdb SauceNAO 35234234.png, 52KiB, 600x800

Anonymous Wed 06 Apr 2022 23:28:10 No.14370406 Report

Quoted By:

>Source?

Anonymous

Anonymous Wed 06 Apr 2022 23:30:17 No.14370414 Report

Quoted By: >>14370450 >>14370477

>>14370405
>Symbolic computer vision and NLP approaches
No one said anything about those. I'm just reminding you that literally no one who actually works with neural networks on a practical level thinks they're suitable for general intelligence.

Anonymous

View Same Google iqdb SauceNAO neuralnet.png, 166KiB, 852x418

Anonymous Wed 06 Apr 2022 23:40:11 No.14370445 Report

Quoted By: >>14370453

>>14370341
>imaginary program
Visual Question Answering already exists.
https://arxiv.org/pdf/1606.00061.pdf

Anonymous

Anonymous Wed 06 Apr 2022 23:42:03 No.14370450 Report

Quoted By: >>14370460

>>14370414
>I'm just reminding you that literally no one who actually works with neural networks on a practical level thinks they're suitable for general intelligence.
https://openai.com/about/

Anonymous

Anonymous Wed 06 Apr 2022 23:42:13 No.14370453 Report

Quoted By:

>>14370445
Absolutely desperate attempt to deflect with irrelevancies and move the goal post.

Anonymous

View Same Google iqdb SauceNAO 1593554619336.png, 78KiB, 536x1515

Anonymous Wed 06 Apr 2022 23:43:25 No.14370459 Report

Quoted By:

>>14368408
You just KNOW /sci/ would fall for this.

Anonymous

Anonymous Wed 06 Apr 2022 23:43:47 No.14370460 Report

Quoted By:

>>14370450
I don't care what their corporate PR statement says. If you had any idea what you're talking about, you'd know what I'm talking about. Just stop posting.

Anonymous

Anonymous Wed 06 Apr 2022 23:47:48 No.14370477 Report

Quoted By: >>14370493

>>14370414
>I'm just reminding you that literally no one who actually works with neural networks on a practical level thinks they're suitable for general intelligence.
https://www.deepmind.com/blog

Anonymous

Anonymous Wed 06 Apr 2022 23:53:17 No.14370493 Report

Quoted By: >>14370548 >>14370567

>>14370477
Don't care. Point stands completely unchallenged.

Anonymous

Anonymous Thu 07 Apr 2022 00:12:50 No.14370548 Report

Quoted By: >>14370586

>>14370493
Is Jurgen Schmidhuber a good enough reference for you? He invented LSTMs.
https://people.idsia.ch/~ciresan/data/agivision2011.pdf

Anonymous

Anonymous Thu 07 Apr 2022 00:18:58 No.14370567 Report

Quoted By: >>14370589 >>14370597

>>14370493
You're already outclassed by large language models. Soon, you'll no longer be employable as a mop-pusher when a roomba powered by a distilled perceiver takes over the remaining task that gives you purpose. You'll either be subsidized or euthanized. I'll be voting for the cheaper option.

Anonymous

Anonymous Thu 07 Apr 2022 00:24:02 No.14370586 Report

Quoted By: >>14370597 >>14370642 >>14370707

>>14370548
Nothing in that paper supports your delusions. Anyway, this is just dull. I don't care about your pop-culture view of the subject. Talk to anyone who's used NNs to solve a real-world problem. Listen to lectures and talks by actual engineers. Let them tell you just how "easy" it is to get a NN to get good at even just one specific task, and how machine learning compares to human learning. Compare how many elements the most complex neural networks have, and how many orders of magnitude more the human brain has, and then add several orders of magnitude more to account for the fact that the nodes and weights in a NN are huge simplifications compared to actual neurons. Then contemplate the fact that playing around with words is actually the most trivial kind of problem for NNs, and that even a task as simple as grabbing an item from a crowded shelf without knocking everything over is currently completely untenable. Your entire comprehension of reality is based on MSM and pop-soi. There is no discussion to be had with you.

Anonymous

Anonymous Thu 07 Apr 2022 00:25:06 No.14370589 Report

Quoted By:

>>14370567
The only "people" outclassed by language models are mouth-breathign imbeciles like you. See >>14370567 and then hang yourself. Your fantasies aren't happening and everyone who's in the know thinks you're a fucking moron.

Anonymous

Anonymous Thu 07 Apr 2022 00:26:49 No.14370597 Report

Quoted By:

>>14370567
The only "people" outclassed by language models are mouth-breathign imbeciles like you. See >>14370586 and then hang yourself. Your fantasies aren't happening and everyone who's in the know thinks you're a fucking moron.

Anonymous

Anonymous Thu 07 Apr 2022 00:28:21 No.14370601 Report

Quoted By:

Once we hit 888 billion parameters, you're going to see some serious shit.

Anonymous

Anonymous Thu 07 Apr 2022 00:30:54 No.14370612 Report

Quoted By:

>A NEW CHATBOT?!

Anonymous

Anonymous Thu 07 Apr 2022 00:31:26 No.14370617 Report

Quoted By:

I JUST PUT ALL MY MONEY INTO AI STOCKS. THIS IS IT

Anonymous

Anonymous Thu 07 Apr 2022 00:34:32 No.14370632 Report

Quoted By:

a more advanced Turing machine was broken (deciphered) about a year ago.. it was a big contest with only one winner.. system was substaintially more advanced than the example here

Anonymous

Anonymous Thu 07 Apr 2022 00:37:10 No.14370642 Report

Quoted By: >>14370648

>>14370586
>even a task as simple as grabbing an item from a crowded shelf without knocking everything over is currently completely untenable.
https://www.youtube.com/watch?v=x4O8pojMF0w

Anonymous

View Same Google iqdb SauceNAO scorecard.jpg, 61KiB, 595x327

Anonymous Thu 07 Apr 2022 00:38:14 No.14370644 Report

Quoted By:

sig-nificance of score? who was involved?

Anonymous

Anonymous Thu 07 Apr 2022 00:39:12 No.14370648 Report

Quoted By: >>14370663

>>14370642
Another pop culture reference, but the point still stands completely unchallenged. Looks like you're a literal NPC. It doesn't look like you understand the concept of grabbing something from a shelf. Am I talking to a literal bot?

Anonymous

Anonymous Thu 07 Apr 2022 00:44:40 No.14370663 Report

Quoted By: >>14370673

>>14370648
https://arxiv.org/abs/1603.02199

Anonymous

Anonymous Thu 07 Apr 2022 00:48:38 No.14370673 Report

Quoted By: >>14370690 >>14370696

>>14370663
Point still stands completely unchallenged. You are a literal NPC, aren't you? You don't comprehend what grabbing something from a crowded shelf means, so you keep posting corporate PR and random references to distantly related things.

Anonymous

Anonymous Thu 07 Apr 2022 00:52:46 No.14370683 Report

Quoted By: >>14370694

How do I get to use this?

Anonymous

Anonymous Thu 07 Apr 2022 00:54:22 No.14370690 Report

Quoted By: >>14370707

>>14370673
A language model based on your posts would be a bipartite tree with two leaves:
>NPC!
>Delusional!

Anonymous

Anonymous Thu 07 Apr 2022 00:56:19 No.14370692 Report

Quoted By:

>>14368408
Does the Turing test actually mean anything objective and significant?

Anonymous

Anonymous Thu 07 Apr 2022 00:57:06 No.14370693 Report

Quoted By: >>14370697

>>14368422
>It's already doing better than human average at these language tasks
Then it has already failed the Turin test. For it to have passed the Turin test it needs to be sufficiently advanced in order to fool a human that it's human. However it's too advanced to be human.

Anonymous

Anonymous Thu 07 Apr 2022 00:57:20 No.14370694 Report

Quoted By:

>>14370683
You don't. They get to use it against you.

Anonymous

Anonymous Thu 07 Apr 2022 00:57:20 No.14370695 Report

Quoted By:

>>14370355
aka make the polynomial longer

Anonymous

Anonymous Thu 07 Apr 2022 00:57:24 No.14370696 Report

Quoted By: >>14370710

>>14370673
So the difference between solving a rubiks cube with one hand and taking something off a shelf is the difference between humans and AGI? Do you think if Boston Dynamics wanted to take something off a crowded shelf, they wouldn't be able to even though their robot can do backflips fine?

Anonymous

Anonymous Thu 07 Apr 2022 00:58:31 No.14370697 Report

Quoted By:

>>14370693
>Turin
Nothin like da Murrican Turrin Test YEE HAAW

Anonymous

Anonymous Thu 07 Apr 2022 01:01:00 No.14370705 Report

Quoted By:

>>14368422
His point is that people will always find an excuse for why it hasn't been solved.

Anonymous

Anonymous Thu 07 Apr 2022 01:01:47 No.14370707 Report

Quoted By: >>14371223

>>14370690
See >>14370586 and then hang yourself. Your inability to grasp what is involved in basic everyday task is irrelevant to my main points, and even your own paper illustrates the problem that it literally took almost a million attempts just to learn a vastly reduced version of a common task. The capacity of a neural network to learn any real-world task is abysmal, and the same pattern repeats in every practical application. Anyway, you're clearly something less than human so this is your last (You). AGI will never exist. The singularity will never happen. The only point of these highly constrained, domain-specific technologies is to manipulate label-thinking mouth breathers like you by playing around with labels.

Anonymous

Anonymous Thu 07 Apr 2022 01:02:01 No.14370708 Report

Quoted By:

>>14368408
Can't wait for the novel AI model. That solo ERP is about to get real.

Anonymous

Anonymous Thu 07 Apr 2022 01:03:28 No.14370710 Report

Quoted By:

>>14370696
You are literally subhuman.

Anonymous

Anonymous Thu 07 Apr 2022 01:04:29 No.14370712 Report

Quoted By:

>>14368422
>It's already doing better than human average at these language tasks
Human average is lowering over time thanks to technology
https://phys.org/news/2013-05-victorian-era-people-intelligent-modern-day-counterparts.html

Anonymous

Anonymous Thu 07 Apr 2022 01:16:50 No.14370739 Report

Quoted By: >>14371006

>>14368408

I tried GPT3, it's not even close to cracking the turing test

Anonymous

Anonymous Thu 07 Apr 2022 02:38:47 No.14370929 Report

Quoted By:

>>14368408
In a weird way the Turing test is already invalidated.
The way we will be able to identify the human is that the human's answers will be dumber than the machines.

Anonymous

Anonymous Thu 07 Apr 2022 02:49:19 No.14370951 Report

Quoted By:

>>14368408
https://www.google.com/amp/s/scitechdaily.com/how-fast-can-electronics-get-quantum-physics-sets-a-speed-limit/amp/

Never because silicon brains will never have the processing power for it.

Anonymous

Anonymous Thu 07 Apr 2022 03:13:23 No.14371006 Report

Quoted By:

>>14370739
This thing seems a lot better than GPT3 though. I don't remember 3 being able to detect anti-jokes.

Anonymous

Anonymous Thu 07 Apr 2022 05:15:21 No.14371212 Report

Quoted By: >>14372965

>>14370307
Can you give an actual example?

Anonymous

Anonymous Thu 07 Apr 2022 05:22:37 No.14371223 Report

Quoted By: >>14371620 >>14372865

>>14370707
>and even your own paper illustrates the problem that it literally took almost a million attempts just to learn a vastly reduced version of a common task.
Are you fucking stupid.
How many attempts do you think it takes for a newborn to do literally anything?

I guess babies don't count as an intelligence according to you?

Anonymous

Anonymous Thu 07 Apr 2022 05:34:07 No.14371240 Report

Quoted By: >>14372906

I for one welcome our A.I. Overlords

Anonymous

Anonymous Thu 07 Apr 2022 11:07:41 No.14371620 Report

Quoted By: >>14372827

>>14371223
>How many attempts do you think it takes for a newborn to do literally anything?
a few tens of thousands at worst (mind you, all this while the brain is taking care of all the non conscious functions of the body)

Anonymous

Anonymous Thu 07 Apr 2022 19:39:51 No.14372827 Report

Quoted By: >>14372870 >>14372904

>>14371620
NTA, but the baby starts off with evolutionary pre-programming and instincts that help it learn efficiently, because it evolved to be in that particular environment. The AI doesn't make any assumptions about what it's doing beforehand, so it makes sense that it would need an order of magnitude more examples. The AI still has some advantages though:
>the model can train itself 24/7, but the baby is idle most of the time
>the model can learn in a simulated environment that runs faster than the real world
>you only have to train the modal once, instead of needing to start over again like with every new human that gets born
The human brain is incredible though. It's not surprising that our best AIs (artificial neural networks) were inspired by biology.

Anonymous

Anonymous Thu 07 Apr 2022 19:53:31 No.14372865 Report

Quoted By:

>>14371223
>How many attempts do you think it takes for a newborn to do literally anything?
Doesn't take literally a million attempts just to learn how to grab something, mouth-breathing imbecile.

Anonymous

Anonymous Thu 07 Apr 2022 19:54:33 No.14372870 Report

Quoted By:

>>14372827
>an order of magnitude more examples
Try 5-6-7 orders of magnitude.

Anonymous

Anonymous Thu 07 Apr 2022 20:02:53 No.14372904 Report

Quoted By:

>>14372827
>NTA, but the baby starts off with evolutio-
*ignored*
I can't believe how many people here were filtered by evolution

Anonymous

Anonymous Thu 07 Apr 2022 20:04:36 No.14372906 Report

Quoted By:

>>14371240
AI is a literal meme. Civilization will collapse by 2050 because of climate change. Good luck working on AIs when we don’t have electricity

Anonymous

Anonymous Thu 07 Apr 2022 20:04:38 No.14372907 Report

Quoted By:

>>14368408
Wow a computer can read data and recite that the color blue is known to be sad? incredible just so incredible I cannot believe this is true
sad = blue
angry = red
NOOOOOOOOO MUH INTELLIGENCE

Anonymous

Anonymous Thu 07 Apr 2022 20:21:08 No.14372940 Report

Quoted By:

>>14368408
In the future the internet is just going to be AIs talking to each other spouting propaganda from a million different companies.

Anonymous

Anonymous Thu 07 Apr 2022 20:29:40 No.14372965 Report

Quoted By:

>>14371212
>Can you give an actual example?
How can a 6 and a 9 overlap to make infinity?

Anonymous

View Same Google iqdb SauceNAO 191826353637.png, 175KiB, 342x342

Anonymous Thu 07 Apr 2022 22:06:44 No.14373302 Report

Quoted By:

>>14370216

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches