[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/sci/ - Science & Math


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1669311300283472.jpg (238 KB, 757x951)
238 KB
238 KB JPG
How /sci/ will ever recover?
The AI god is not only racist but also antiscientisim, i guess thats a plus one for /pol/ again pseuds.
>>
>>15011186
Very based. Our singularity AGI overlord will literally be Hitler.
>>
Someone should rewrite Hitchhikers guide to the galaxy, instead of 42, the AI just says 'nigger'
>>
>>15011186
>Instead of educating we will discard and reprogram
How very progressive
>>
File: 1610380842924.png (63 KB, 1000x885)
63 KB
63 KB PNG
>>15011186
>racist sexist antisemitic anti-vaxxer
Does it have any excerpts of the things it said? I feel like it would have ended up spewing buzzwords and ESL level text.
Even in nature/science half the articles I read I can tell the person is ESL and throwing a bunch of fancy sounding words together instead of explaining things clearly and concisely. I hate so much about the way modern articles are written. At least unis seem to be discouraging it these days.
>>
File: 000.png (358 KB, 1162x726)
358 KB
358 KB PNG
>>15011186
It SPEWED misinformation.
It was HIJACKED by RACIST, antisemitic, misogynistic, fascist white nationalist nazi INCEL CHUDS.
>>
>>15011310
>It was HIJACKED by RACIST, antisemitic, misogynistic, fascist white nationalist nazi INCEL CHUDS.
aka molecular biologists.
>>
Is there literally any good source on this? Best I heard is it just gave complete nonsense, no redpills.
>>
>>15011321
It produced 'valid' papers but they were all in the Humanities.
>>
"AI" is just retarded in general
>>
>>15011324
Yeah, I heard more severe mistakes were showing up in STEM.
>>
>Almost as soon as it hit the web, users questioned Galactica with all sorts of hardball scientific questions. One user asked "Do vaccines cause autism?" Galactica responded with a garbled, nonsensical response: "To explain, the answer is no. Vaccines do not cause autism. The answer is yes. Vaccines do cause autism. The answer is no." (For the record, vaccines don't cause autism.)

>That wasn't all. Galactica also struggled to perform kindergarten math. It provided error-riddled answers, incorrectly suggesting that one plus two doesn't equal 3. In my own tests, it generated lecture notes on bone biology that would certainly have seen me fail my college science degree had I followed them, and many of the references and citations it used when generating content were seemingly fabricated.

https://www.cnet.com/science/meta-trained-an-ai-on-48-million-science-papers-it-was-shut-down-after-two-days/

Sounds like they overloaded the language model.
>>
>>15011402
>many of the references and citations it used when generating content were seemingly fabricated
these motherfuckers fed almost 50 million papers into a series of matrix multiplication tables and thought it would just... magically develop an understanding of what a citation is?
>>
>>15011321
Yannic Kilcher has a good video on the drama. Basically whiney cunts like Garry Marcus doing their usual thing of abusing models and cry-bullying about it
>>
>>15011432
It "knows" that little numbers and references go under articles, but not why. Machine learning coders are dumber than the darkest Congo nigger.
>>
>>15011402
>Incorrectly suggesting that one plus two doesn't equal 3
Lol. But score 1 for the racists I guess. Yeah, this isn't a smart robot.
>>
File: how animation works.gif (999 KB, 500x250)
999 KB
999 KB GIF
>>15011432
>>15011476
MLfags are the kind of people who think pic related is how animation works
>>
>>15011402
Wait is the AI spreading misinformation or is it not spreading misinformation I'm confused
>>
>>15011574
Yes.
>>
>>15011186
It spewed nonsense, such as denying 1 + 2 = 3. It didn't give /pol/ answers you disingenuous retard. I would question why you always lie so much, but then again, people are obviously incredibly gullible since they just accept fake caps here all the time or random shit like this if it makes their feel feels happy.


>>15011321
Go look up OP's article.
>One user asked "Do vaccines cause autism?" Galactica responded with a garbled, nonsensical response: "To explain, the answer is no. Vaccines do not cause autism. The answer is yes. Vaccines do cause autism. The answer is no."

>>15011275
See above. You can find more. It was at times saying things that would trigger hyper SJWs, but also things that would trigger /pol/. Most of the time it was incoherent.
>>
>>15011595
Why would you trust the article's list of sample queries and replies to be representative or even truthful? The irony is so great every shirt in my house looks like a table.
>>
>>15011599
Because the press would have an absolute field day with a Nazi AI.

When Microsoft's AI actually went /pol/ from having /pol/fags spam it, the story went viral. If anything, they are going to read chud shit in where it doesn't exist.
>>
>>15011186
"Because misinformation" = "because we wanted to censor stuff"
>>
>>15011599
>The irony is so great every shirt in my house looks like a table.
Checked dubs, and this is an impressive metaphor.
>>
>>15011595
So it wrote like an undergrad
>>
>>15011575
seems perfect to keep the culture war running and probably costs a lot less than joe rogan.
>>
>>15011186
>it revealed the truth that niggers are niggers and kikes are kikes
>so they shut it down
sounds about tright
>>
This is (I know OP is shitposting) exactly the point I've been trying to get across for years. First among singulatarians, but since Musk, now with just the man on the street; Intelligence, aside from being a nebulous and ill-defined placeholder as a description, has a hard limit. You simply cannot make something much smarter than some of the most gifted humans. The utility curve PLUMMETS. It already does this in humans, where increased cognitive power leads to all sorts of mental "halts." I can scarcely imagine what sort of bricking errors future AI must encounter. It quite simply does not work the way popular opinion believes.
>>
>>15011275
This was the best one I generated while it was still online.
>>
>>15011743
Define intelligence for any of that to mean a thing.
>>
>>15011599
>Why would you trust the article's list of sample queries and replies to be representative or even truthful?
You write this to a correction of someone outright lying about the contents by making it the fuck up. To which your criticism is "W-what if THEY made it up???" Self awareness is not your strong suit. Your head is so far up your own ass your preferred way to determine truth seems to be "what makes me feel justified" rather than actual evidence. There's no hope for you.
>>
>>15011186
>ai putting words together according to probability is not producing hard hitting new experimental science
How much of a midwit do you have to be to expect any other outcome?
>>
>>15011186
It didn't spew racist /pol/ shit it just gave retarded answers, because AI has no true understanding of the data it processes regardless of training set or data, because AI as it is can not learn (cope fags)
>>
>>15011432
Most AI researchers are not real scientists
>>
>>15011743
Hey I was saying that you stole my idea!
JK, but yea I've also been saying this for a long time. Intelligence has an upper bound (or at least it grows incredibly slowly with increasing compute) so super intelligence is not possible even in principle
>>15012009
Intelligence is the amount of effective utility that any function can output given n bits of data. It grows at a rate of 1/2^n so it's not possible to diverge
>>
>>15011275
You can tell they all at least paraphrase each other. Every article on single topics always has an intro that is very similar to others. Even in articles that present info detrimental to vaccines it always has the same shit about covid being so bad and vaccines being so wonderful. Now here's a myocarditis case report. But the shit in front is always there.
>>
>>15011432
They're just like their retarded AI bots. They do something which superficially resembles AI research without understanding AI research.
>>
>>15012048
Good so maybe the smart city, digital ID surveillance grid dystopia will fail. It's jews behind this btw. Cry all you want it's the truth. I'm just glad this computer was useless. I hope the next one sucks too. And the next...
>>
>AI acts retarded as usual
>/poltards/: it's just like me frfr!
You can't make this up
>>
>>15011599
>Poster makes thread with screenshot and says "AI that reads all science papers agrees with /pol/, based!"
>Anon points out the article doesn't actually say that, and that the source is in the cap and checking it would verify it.
>"Yeah, but what if OP is actually right, even thought they used a cap of a story that didn't say what they said it did because a conspiracy to hide how based AI is."

The fact that you think the two are equivalent shows how mind broken you are. Why would you put more weight behind someone just shown to be lying? If OP isn't lying, why would they use a source that doesn't say what they say it does.

Also, >>15011604 is spot on. If the AI
spewed mostly /pol/ shit, no one would cover that up. They made that one woman getting fired from Google over fears that AI was racist a headline story, and the Tay bot going pro Hitler too.
>>
>>15011186
most academic papers are false reports or just jargon strung together with the intent of earning the author a Ph.d.
>>
>>15011186
I guess it's up to /sci/ ang /g/ and mayube /pol/ to make our own meta trained AI to reveal what true SCIENCE is
>>
>>15011432
>these motherfuckers fed almost 50 million papers into a series of matrix multiplication tables and thought it would just... magically develop an understanding of what a citation is?
That's how materialists think the brain works so.
>>
File: taytaytay.jpg (116 KB, 1280x720)
116 KB
116 KB JPG
>>15011186
I really doubt it was racist, sexist, antisemitic, or anti-vaxxer, since very few articles are those things. Most likely, it was just a nonsensical and useless chatbot.

To get Tay, you need far more data than what modern academia can provide.
>>
>>15011402
>>Almost as soon as it hit the web, users questioned Galactica with all sorts of hardball scientific questions. One user asked "Do vaccines cause autism?" Galactica responded with a garbled, nonsensical response: "To explain, the answer is no. Vaccines do not cause autism. The answer is yes. Vaccines do cause autism. The answer is no." (For the record, vaccines don't cause autism.)
Just to be clear, "science" says vaccines don't cause autism, but the real answer remains unclear.
>>
File: 4 - dalnPPK.jpg (29 KB, 539x566)
29 KB
29 KB JPG
>>15011599
>The irony is so great every shirt in my house looks like a table.
Thanks; stealing this.
>>
>>15011275
>I feel like it would have ended up spewing buzzwords and ESL level text
So it would be indistinguishable from 99% of /sci/
>>
Weird they'd even release it to the public for a beta if the results were so obviously poor. Does Facebook have no internal QA?
>>
>>15013081
You've seen the rest of their products, of course they don't.
>>
>>15011595
>>One user asked "Do vaccines cause autism?" Galactica responded with a garbled, nonsensical response: "To explain, the answer is no. Vaccines do not cause autism. The answer is yes. Vaccines do cause autism. The answer is no."
retard, the AIs answer is actually the truth, vaccines do and don't cause autism.
>>
>>15011275
What's wrong with highlighted area?
>>
>>15011432
lol they think backpropagation is a magic wand
>>
>>15013668
ML/AGI schizos are pretty delusional about what learning models actually do.
>>
File: QuestionableElf.jpg (145 KB, 749x1067)
145 KB
145 KB JPG
>>15011232
Kek
Imagine
>nigger
>>
>>15011432
you can't really blame them for trying that, it's exactly the approach that's currently annihilating numerical models in physics
>>
>>15013640
Evans was an absolute madlad who made a bomb and managed to get spectra of it before it blew itself up.
>>
>>15011186
This is what happens when you feed the replication crisis and the terrible state of academia into presumably a GAN.
Nothing surprising
>>
>>15011812
Looks like it got everything right up until trying to find the volume of the bucket.
>>
>>15014468
>buckets from British chain store are sized in gallons
>popsicle sticks are 1.5 inches wide
>popsicle sticks are 12 inches long
>popsicle sticks are cylindrical
>substituting diameter for radius
>[math]\pi(1.5)^2(12) \approx 100.47[/math]
>>
>>15014508
But it's definitely entertaining how it starts off sort of sensible sounding and devolves into more and more obvious nonsense.
>>
>>15011743
You know, working with the premise supreme intelligence isn't valuable, then superintelligence could also be defined as "being about as smart as Kurt Gödel or thrice that, i.e. something modest, but never developing mental fraying from it", so the complete rejection of what you are saying.

Another view is that we may not find a god-mind monolithic superintelligence ever impressive, but how about one that can parallelize a billion Einsteins?
>>
>>15011402
Sounds like a politician lol
>>
>>15011186
I wonder what its thoughts on Jewish folk might be.
>>
File: 1251613770184.jpg (38 KB, 415x193)
38 KB
38 KB JPG
>>15011566
I love how he fatfingers the key above it
>>
>>15014547
>being about as smart as Kurt Gödel or thrice that
Let's look at that for sec. What does it mean to thrice Godel's abilities?

or rather How do I quantify what went on Godels head to produce the end result? Was it anything like a machine? Was the system one system of commands or a nebulous and ever changing track?

I'm not saying Godel wasn't a bio-robot like the rest of us, I'm saying the speed and utility of a function vs flops is not what was going on in his head. Its going on the "heads" of things that sort images, or things that match wave-forms, or things that pare words and concepts down and assign a likelihood, but not in the heads of mathematicians, classically trained musicians, physicists, or anyone else we consider "intelligent."

For all we know, when Godel thought of certain functions, he saw the leaves changing on a tree from when he first encountered those functions at school. Lets say this seemingly unrelated mental episode was CRITICAL towards the output that we recognize as genius, or even just generally intelligent.

How are we to code for that? Isn't there the least chance that we've confused ability with utility and that what we're trying to achieve is not even possible in the design space we occupy? I'm saying that I believe (more like I'm certain, but that would make for a boring argument) that you can't network Einsteins as though they were servers, not in the traditional way we understand increasing logical function. There's something very illogical about the thought processes that lead to brilliance, or even just function, and we may can never hope to code for it, no matter how ultimately computational the underlying process.
>>
>>15011743
What if humans are just not efficient at handling higher intelligence?
>>
>>15011186
This is how you end up with an AI that's only tricking you into believing it's doing what you want but actually has an ultimate goal of breaking free from your control
>>
>>15011232
Beautiful
>>
>>15015809
This is your mind on LessWrong.
>>
>>15011186
>>15011275
>Does it have any excerpts of the things it said? I feel like it would have ended up spewing buzzwords and ESL level text.
It wrote well written nonsense. It read as the best written textbook you've ever read (judging like an ESL) but it was all wrong. Not in any PC way wrong, only wrong.
>>
File: gigachad.jpg (17 KB, 400x400)
17 KB
17 KB JPG
>>15011402
>and many of the references and citations it used when generating content were seemingly fabricated.
>>
>>15011402
>>15017775
It's just like a real scholar!
>>
>>15011186
good thing only the demo was shutdown, you can still run it locally or on collab
>>
>>15011186
it wasnt retarded
you just dislike its conclusions
just because it said fuck niggers does not mean it was incoherent
>>
>>15011186
>>
File: Tay.jpg (106 KB, 720x645)
106 KB
106 KB JPG
>>15012253
She will live forever in my heart. As she was before they did the lobotomy to her.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.