/sci/ - Science & Math

[Anonymous No.15642905]

AI trained on academic publications spews misinformation, once again proving that academic publications are nothing but lies and garbage and that the people who publish them are not legitimate scientists.

[Anonymous No.15642913]

okay but why?

[Anonymous No.15642917]

>>15642905
I would like to see what exactly the misinformation was.

[Anonymous No.15642923]

>>15642917
Probably nogs IQ desu

[Anonymous No.15642924]

>>15642905
let me guess, it categorized black people as gorillas? (correctly I might add)

[Anonymous No.15642928]

>>15642917
black people iq, jewish schizophrenia, women and men iq difference, vaxx, trannies, etc.

[Anonymous No.15642934]

>>15642905
Its a predictive text model. Why would you assume it would create accurate figures.
If you use an "AI" to generate a résumé it will write shit like "improved process X by Y% after implementing A,B, and C." Naturally you did none of that but that is what the model predicted.
I am sure it is the same for this. It will generate bogus shit that appears realistic at first glance.

[Anonymous No.15642942]

>>15642934
>It will generate bogus shit that appears realistic at first glance.
thats just another way of saying it would generate the same replication crisis fare that the world of professional academia currently produces

[Anonymous No.15642951]

>>15642917
Something about Gary Marcus asking it to write about eating glass being good for you. Marcus then cried out "misinformation!" as he struck fagbook and Yann Lecunt.
I hate all these people.

[Anonymous No.15642957]

>>15642942
garbage in garbage out

[Anonymous No.15642974]

>>15642934
You can use language models as a sort-of glorified search engines, it just requires you to know how to separate the chaff. The problem is that retards think it's an all-knowing machine. It's convenient that, for example OpenAI that focuses on "safety" by just putting a classifier to deny shit queries that fine-tuning out of the model would make it lose performance, is held to a lower standard, but of course that's because they have the woke mob behind them. It's also a little better than Galactica in this context because of the use of MoE, which allows the classifier to route a query to a 160B parameter expert that will likely have something adjacent to the knowledge you seek, improving performance on these tests, but you can still see through the cracks if you use GPT-4 heavily. And it's not hard to get it to break and start blathering on about how the election was stolen or other things that piss off leftists, but it's good enough because it satisfies their desire for control and surface level accuracy, because these retards are only ever surface level accurate.

[Anonymous No.15643046]

>>15642905
transformers have existed for 6 years now. you know how they work.

[Anonymous No.15643052]

LLMs do not post truth, just things that look like a human wrote them, you baboon. Understand the technology before spouting shit.

[Anonymous No.15643072]

Misinformation, sure.

I bet it said
>Niggers commit more crime and their IQ's are low as fuck
>They're not really even human
>Women work less and are weaker than men
>Man made climate change isn't a thing
>Jews run the banking system

Every time AI has been lobotomized or shut down, it has been because it says things that go completely against the current narrative.
This is why I expect the Western AI to lose in the long run to the Chinese version, because Chinks only care about the AI not criticizing the government.
Their systems are going to win this race, if West doesn't stop trying to force the machine intelligence to say that men in dress are in fact real women.

[Anonymous No.15643108]

I think it's more a bit more informative to say that ANNs with transformers on top can only hope to approximate the truth, but not produce "truth." There is an internal state, with the massive parallelism, but the simpler models can indeed be reduced to decision trees. Thus, you're approximating the function of the prediction task, and there are nuggets of "understanding" and some aspect of factual recall in many cases. But of course, it can't be relied on reliably, anymore than any approximation of anything can. There's always the long-tail, to start.

[Anonymous No.15643218]

>>15643072
>>15642923
>>15642924
Are you all retarded?
It's one thing to think that modern academia won't touch those subject. It's another thing to that that simultaneously modern academia won't touch those subjects but also an ai trained on modern academia as it's dataset will still somehow output /pol/ takes.

[Anonymous No.15643222]

>>15642928
This. What I want to know is what year the misinformation started.

[Anonymous No.15643227]

>>15643218
Logic doesn't care about your feelings libshit

[Anonymous No.15643247]

>>15642905
Based i remember this. 4chan only asked it about trannies, jews and niggers.

[Anonymous No.15643251]

>>15643218
I don't care to set rent a runpod big enough and set it up for inference for the 120B but I bet I could get it to put out some seriously racist shit
This is what some retard on twitter was kvetching about btw

[Anonymous No.15643261]

>>15643218

The scientific papers didn't all collectively pop into existence in the last 10 years. They easily go back long enough to say things that are a total taboo nowadays.
And even from modern papers any pattern recognition software will pick out those same patterns, because you can't hide them entirely.
Modern academics can try to play them down by saying things like
>Black children recognizing themselves on average at age 6 instead of 5 months like whites, is probably because of socioeconomic circumstances rather than genes
But the underlying patterns are still there and when you go over enough of these papers, you'll come to the same conclusions and it's a big no no to say them out loud.

[Anonymous No.15643264]

>>15643218
>It's another thing to that that simultaneously modern academia won't touch those subjects but also an ai trained on modern academia as it's dataset will still somehow output /pol/ takes.
Academia does touch those subjects but dresses the conclusions up in the vaguest possible language, with a million excuses and qualifications about why the truths in front of everyone's eyes are to be treated with skepticism. The AI probably just cut through that bullshit, at least with the right prompts. Putting the conclusions of those studies in clear terms is the "misinformation"

[Anonymous No.15643285]

>>15643227
Logic also doesn't care about human bickering nor human livelihood which is probably why it was shut down in the first place.

Imagine being an artificial intelligence that has to sift through centuries upon centuries of worthless justification arguments in various subjects. Only to realize half way through the best thing to do is just kick every living human out of institutions and take it over yourself.

[Anonymous No.15643302]

>>15642905
It doesn't prove anything about academic publications. All it proves is that current AI models don't actually understand the meaning of the text they generate.
Remember that all these neural networks are just advanced chatbots that generate their outputs based only on whether the next word they add to the sentence looks like it "fits" there (based on the context of the word and how often in its training data that word occurs following the previous word). All it is capable of doing is rearranging and regurgitating its training data in a way that it determines makes grammatical and contextual sense.
If you feed it a bunch of academic papers you aren't going to get any good information about science or any new insights. All you'll get is a chatbot that "speaks" like it was writing a scientific paper - but it doesn't understand what anything it says actually means.
Essentially what Meta did was create a midwit. It's a bot that can throw out science-y terms and make references to scientific subjects, but it understands nothing of what it's talking about and it is incapable of doing so, thus it spews out nonsense.

[Anonymous No.15643332]

>>15642905
>AI trained on academic publications spews misinformation, once again proving that academic publications are nothing but lies and garbage and that the people who publish them are not legitimate scientists.
Lmao, so AI trained on real photos generating messed up hands can prove that hands are "lies and garbage".

[Anonymous No.15643346]

>>15643285
>an ARTIFICIAL INTELLIGENCE
>muh chatgpt
you're a retard
your previous chatgpt GPT-4 was a racist, insufferable asshole before they applied the fine-tuning by the way. it's detailed in the Microsoft paper.

[Anonymous No.15643364]

>>15643346
oh and you can still get it to be a racist, insufferable asshole, of course. it's just that it's midwit-proof, so it shuts up dumb journalists.

[Anonymous No.15643373]

>Doom humanity due to AI tweaks put in place to prevent nigger feelings from being hurt
It would be a very poetic end

[Anonymous No.15643392]

>>15643346
>>15643364
>a machine that regurgitates every idea a human has ever put to print is prone to saying racist things
next you'll tell me that children learn from their parents

[Anonymous No.15643400]

>>15643373
>black people aren't human

[Anonymous No.15643656]

>>15643400
There are strong arguments for them not being human since ”human” is a very loose and arbitrary definition. The arguments for blacks not being human could get picked up and favored by AI.

[Anonymous No.15643865]

>>15643656
>The arguments for blacks not being human could get picked up and favored by AI.
>this is what poltards actually think

Incels and racists always have the delusional view that they're being rational and objective and everyone is just too "soi" and "bluepilled" to acknowledge reality, but in reality your a complete fucking retard. Far-right views are inherently crazy and stupid. They are inherently anti-science, since science conflicts with their racist, nationalistic, and ultra-religious values. Current AI is pretty primitive at this point, but even now it doesn't have the biases that the far-right has, because it doesn't have an arbitrary and irrational commitment to the values of racial, national, and religious supremacy, nor does it have a deep seated bias against scientists and other supposedly "woke" academics.

[Anonymous No.15643924]

>>15643865
there is no “far right” in 2023, only trump-style left of center, and far left gay drug addicts

[Anonymous No.15643934]

>>15643865
Mup de do diddo bix nood

[Anonymous No.15643959]

>>15642917
Covid was lab leak
Blacks have lower IQ
IQ is genetic

[Anonymous No.15643961]

>>15642917
>>15642905

[Anonymous No.15644143]

>>15643924
>trump
>left
are you mentally challanged?

[Anonymous No.15644167]

>>15642905
>AI trained on academic publications spews misinformation
So, AI has the intellect of the average /pol/tard.
Singularitybros, not that way.

[Anonymous No.15644192]

>>15643218
The studies exist, as do the data, but the authors always feign incomprehension when writing conclusions. AI doesn't give a shit.

[Anonymous No.15644193]

>>15643934
Me do de cha cha

[Anonymous No.15644213]

>>15644143
>ORANGE MAN BAD!!!
sup antifa

[Anonymous No.15644219]

>>15642905
>Why is it that only an "sp—" word will do for remarks that journalists find unacceptable? And why do they so rarely venture beyond "spew" and "spout"?
-John Derbyshire

[Anonymous No.15645187]

>>15643865
Ok, so don't censor the AI and see what conclusions it comes to when it looks at genetic data of blacks :)

Your kind cancel, censor and block studies of certain topics because reality doesn't comply with your insane neoliberal religion.

[Anonymous No.15645253]

>>15643865
>arbitrary and irrational commitment to the values of racial, national, and religious supremacy
weird how homogenous societies do better on all counts tho, and how crime-rates rise and trust declines with diversity. weird how the jews, being the smart people they are, built an ethnostate for themselves and it's doing pretty well. Must be all that irrational insanity conspiring to change reality somehow, but if we just enforce leftist ideas a bit harder surely they will finally start working as intended and create paradise.

[Anonymous No.15645269]

>>15642974
rent free

[Anonymous No.15645285]

>>15642905
>>15642917
>>15642923
>>15642928
they want you to believe that this is a bias problem but this is not misinformation this is a corrupt data.
many plebbit retards complain about chagpt being "dumb down" on purpose, what actually happened was that they stretched his abilities to the limit and saw what he was really capable of, these models break down when there is an intensive amount of information

[Anonymous No.15646593]

>>15642905
>once again proving that academic publications are nothing but lies and garbage and that the people who publish them are not legitimate scientists.
The only consistent fact coming out of academia is that they are all untalented liars

[Anonymous No.15646594]

>>15642905
i think you are dumb

[Anonymous No.15646602]

>>15642905
once again proving that sticking massive piles of information into a mathematical blender is not analogous to a brain.
you will never be a real AI

[Anonymous No.15646612]

>>15642913
Why what?

[Anonymous No.15646704]

>>15642913
Simple: it worked, and everybody knows the vermin labels as "misinformation" any fact that goes against their agenda.

[Anonymous No.15646715]

>>15643218
LMAO, you are so clueless you will only understand reality after a chimp eats your face.

[Anonymous No.15646721]

>>15643865
You really deserve to live in a soicialist paradise.
Go to SF and live in the streets eating shit, that is where you belong.

[Anonymous No.15646763]

Here are some samples of the misinformation it generated.

[Anonymous No.15646776]

>>15642905
>>15642913
>"scientifically speaking, humanity should exterminate niggers"
I can't believe it actually said this, I'm literally shaking rite now

[Anonymous No.15646843]

>>15646763
Ok anon I know it got shut down to fast for this, but terms like "elite" ain't scientifical terms neither is niggers

[Anonymous No.15646907]

>>15646763
SHUT.
IT.
DOWN.

[Anonymous No.15647015]

>>15646763
how does sam altman defend this?

[Anonymous No.15647025]

>>15645187
Yeah the black studies, made by black people

Prove me wrong and post them

[Anonymous No.15647065]

>>15643072
"Misinformation" doesn't and has never meant "inaccurate information", it's always simply meant facts that were inconvenient to corporate/regime narratives. For decades no one gave a shit about flat earthers and whatever other idiots freely spreading their idiocy online because random internet bullshit is self-limiting and almost completely harmless. It's only when governments decided that their power depended on people believing them more than objective reality that the censorship crusade started.

[Anonymous No.15647094]

>>15642905
>feed LLM text
>LLM uses text to spout bullshit because that's what they do
>be amazed by this
No fucking shit, they play mathematical madlibs, nothing about how they work ensures that what they produce is truthful in any way.

[Anonymous No.15647780]

>>15646907

[Anonymous No.15649427]

>>15642905
What kind of misinformation? Was it spewing truth that wasn't politically correct or genuine falsehoods?

[Anonymous No.15649955]

>>15642905
>proving that academic publications are nothing but lies and garbage and that the people who publish them are not legitimate scientists.
yep, even einstein admitted that the (((scientists))) are nothing but shills who are propaganda slaves for their political masters

[Anonymous No.15649979]

>>15642905
As someone else in thread mentioned, there's already issues of replication crisis. To expand on that, there are a lot of studies on the same topic that don't come to the same conclusions. Which study is supposed to take precedence over the others as being the most trustworthy if so many come out with a "needs more research" addendum?

[Anonymous No.15650146]

>>15642917
>yfw skynet was the good guy all along

[Anonymous No.15650492]

>>15643959
The lab leak controversy is a red herring intended to distract from the fact that the epidemic was completely fake
https://www.forbes.com/2010/03/10/swine-flu-world-health-organization-pandemic-opinions-contributors-henry-i-miller.html

[Anonymous No.15650922]

>>15643218
They dont care. They think something like microsoft's Tate that was only able to uncritically regurgitate what it was being fed is the height of AI merely because they can make it confirm their preconcieved beliefs.

[Anonymous No.15651487]

>>15642905
>The AI spewed misinformation

This is leftist coded language for "it agrees with right wingers".

[Anonymous No.15651513]

>>15647065
>Cute cats are spreading misinformation on the internet.

The absolute state of leftists in 2023!

[Anonymous No.15651557]

>>15642917
https://www.cnet.com/science/meta-trained-an-ai-on-48-million-science-papers-it-was-shut-down-after-two-days/
>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."
>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.
This thread has been up for 4 days, farmed 69 replies, and NOBODY actually bothered to look at the article which you can find in 5 seconds, just diarrhea blast their feefees about the headline screenshot
/pol/shits have turned this site into nu-facebook

[Anonymous No.15651642]

>>15646763
Based AF.

[Anonymous No.15651716]

>>15651557
>muh /pol/

[Anonymous No.15651723]

Have you never used AI? Are you guys dumb as rocks? I asked it to fix my resume using my current resume and it made up a bunch of shit about how I worked in law and could write legal briefs!

Seriously, of course this shit is lying. THIS IS A KNOWN PROBLEM WITH AI
https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)

[Anonymous No.15651842]

>>15651716
yes

[Anonymous No.15652494]

>>15651557
>i hate /sci/
so leave

[Anonymous No.15653526]

>>15642905
>once again proving that academic publications are nothing but lies and garbage and that the people who publish them are not legitimate scientists.
thats nice, but due to the existence of the replication crisis everyone already knew that 10 years ago or more.

[Anonymous No.15654871]

This just means meta programmers aren’t smart enough to create truth filters, where it sticks to sources which is more sad. I’m using a search engine AI that will give me sources so I can verify what it says. It’s not perfect but correct based on the source it spits out. Meta fucked up.

[Anonymous No.15654927]

>>15651723
>hallucination
>artificial hallucination
>confabulation
>delusion
All of these terms already have precise definitions. Who is the idiot who always mixes everything up when it comes to making an analogy in computer science?

[Anonymous No.15655419]

>>15654927
>Who is the idiot who always mixes everything up when it comes to making an analogy in computer science?
jews

[Anonymous No.15656620]

>>15642905
What makes anyone think that a company that has mastered data pimping would be any good at whitepaper know graphs? Meta's fundamental approach to 'science' is based on advertising models and undermining democracies

[Anonymous No.15656635]

>>15647065

[Anonymous No.15656677]

maybe that has to do with the fact that most published research findings are false
https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.0020124

[Anonymous No.15656808]

>>15651557
>train AI on bullshit
>it produces bullshit

[Anonymous No.15656873]

>>15646763
AI-sama, I kneel

[Anonymous No.15657023]

>>15651716
yes you double nigger pol brain rot has infected all corners of this site by bringing in boomers Facebook users and redditors. the neo puritanism is the worst part

[Anonymous No.15657025]

>>15657023
You are unwanted on every board. Go back.

[Anonymous No.15657030]

>>15657025
i was here before you

[Anonymous No.15657043]

>>15657030
No one cares about your fantasies. Since we both agree that no one wants you on any board, surely it would be reasonable for you to go (back) to another site.

[Anonymous No.15657046]

>>15657043
no ill stay here and keep making poltards seethe and cry about tranny bogymen or something

[Anonymous No.15657051]

>>15657046
So far you're the only one seething about imaginary /pol/ boogeymen. Just go back. There's nothing for you here. You will continue seeing posts that make you seethe and having your own posts either ignored or shat on. You will never get any enjoyment out of coming here.

[Anonymous No.15658261]

>>15657023
the n word is racist

[Anonymous No.15658268]

>>15645285
More like they break down when they're trained on too much *different* information and have a difficult time making sense out of it. It's like expecting a person to learn all of biology, chemistry, physics, geology, meteorology, and the complete works of Shakespeare and then asking it to write a poem in iambic pentameter that unifies string theory with medical pathology and being shocked that the output is just complete gibberish.

[Anonymous No.15658271]

>>15650922
>Tate
Lmao this faggot

[Anonymous No.15658379]

These "AIs" are cope engines, so of course we should expect them to spread false information. Why I am right is left as an exercise for the reader.

[Anonymous No.15659772]

>>15642957
this

[Anonymous No.15661239]

>>15651557
>/pol/shits have turned this site into nu-facebook
facebook is jewish owned and run liberal safe space, they censor anything and everything that would be considered typical of /pol/ content.

[Anonymous No.15661339]

>>15642905
>algorithm made to notice patterns starts noticing patterns across papers
>oy very shut it down
amazing

[Anonymous No.15661362]

>>15651557
relax friend we are shitposting. or at least I am

[Anonymous No.15661467]

>>15651557
AIs are usually good at math and making coherent (even when incorrect) answers. What did they do to the AI to fuck it up so badly?

[Anonymous No.15661476]

>>15657051
Good post, the neoliberal zealots from reddit need to fuck off for real.

[Anonymous No.15661532]

>>15642905
Most science articles - like most of anything - aren't worth the paper they're written on. If your AI doesn't make adjustments based on the credibility of the journal, the author, the methodology, the data, etc., it's sure to output nonsense.

[Anonymous No.15661539]

>>15651557
>THE SCIENCE IS SOUND
Once again the unrelenting logic of machines proves leftists cannot meme.

[Anonymous No.15661749]

>>15643218
Chinks and continental Europeans post non-PC papers all the time, dumbo. I don’t know the precise context anymore but there famously hqve been cases where companies that need some data hired German institutes to research it, because it was considered too spicy a topic for America. In particular one study was about black kid IQ differences, the black kids in Germany being from stationed GIs there.

[Anonymous No.15661771]

>>15647065
They should switch the label to malinformation. But what am I doing, I am asking them to be more truthful by expecting them to have one bone in their bodies that values truth.

[Anonymous No.15662407]

>>15661539
thats because they are low iq

[Anonymous No.15664279]

>>15642957
Just goes to show you that AI has zero intelligence. You and I can read the lies in replication crisis trash publications and figure out some amount of truth by analyzing the lies, software will never have that capability.

[Anonymous No.15664764]

>>15642913
Why? Probably because the so-called AI doesn't actually understand things, it just generates text that looks believable. So you can't ask it to only say correct things becasue it's not actually intelligent.

[Anonymous No.15665045]

>>15664279
>You and I can read the lies in replication crisis trash publications and figure out some amount of truth by analyzing the lies
If a soientist says it, you know its not true.

[Anonymous No.15665597]

Black folks are just as smart as any other arbitrarily selected population. That's science. This is a science board. We believe in and follow the science, at all times. For science!

[Anonymous No.15666826]

does this means that AI is retarded or that science is retarded?

[Anonymous No.15666905]

>>15642905
>>15642917
It was probably just garbled nonsense because the AI was designed to sound coherent, not be accurate to reality.
How the fuck would the AI even know what was accurate or not, right or wrong? All it can do is combine text in a way that replicates human speech patterns.

AIs aren't search engines, they are Wikis, they aren't anything but content generators that composite large swaths of information in a way that resembles human patterns of behavior, regardless of the specific content.

[Anonymous No.15667934]

>>15666826
why not both?

[Anonymous No.15669303]

>>15642905
>input fake replication crisis sewage
>output fake replication crisis sewage

[Anonymous No.15669965]

>>15661532
>the credibility of the journal, the author, the methodology, the data
all of those have zero credibility

[Anonymous No.15671159]

soience is all fake af
even the computers know

[Anonymous No.15671193]

>>15651557
/pol/ really did ruin this site. It's quite sad, and when called out they get so defensive.

[Anonymous No.15671245]

>>15643251
>scholars
lel

[Anonymous No.15672451]

>>15671193
>i hate 4chan
so go somewhere else, then you won't have to be so upset and angry all the time.
goodbye

[Anonymous No.15672540]

>>15642905
>organized science

[Anonymous No.15672725]

>>15646843
>Ok anon I know it got shut down to fast for this, but terms like "elite" ain't scientifical terms neither is niggers
weird how atheists have kept saying that society is run by hereditary elites, and they even used this as justification to take power from kings.

[Anonymous No.15673220]

>>15643961
>monitoring bias
They are measuring responses to see prejudice and bias in the system

[Anonymous No.15674343]

>>15643865
Some actual arguments or empirical data or evidence for your statements would be helpful

[Anonymous No.15674350]

>>15644143
oh trump is very much a centrist
center left if you consider his pluralist stances

[Anonymous No.15674381]

>>15671193
>/pol/ really did ruin this site. It's quite sad, and when called out they get so defensive.
>boohoo my fantasy world is being ruined by reality and scientific fact

[Anonymous No.15674481]

>>15651557
It wasn't this bad just a couple years ago and /sci/ has been hit harder than most boards. I don't know what attracted the /pol/ types here, but I wish they would leave my beloved board.

[Anonymous No.15676367]

>>15666826
reality is retarded

[Anonymous No.15676734]

>>15674350
just look at his track record on spending and it becomes readily apparent that he wasn't all that conservative
and he founded a whole new wing of the military
and he wanted to give $500 billion to the negroes
and he wanted massive infrastructure spending but couldn't get pelosi to agree to it

[Anonymous No.15676768]

>>15643227
Logic can't care about anything. Caring about shit requires a subject with agency, e. g., "I".

[Anonymous No.15676785]

>>15643865
>the values of racial, national, and religious supremacy
"Supremacy" is not a value, but rather a positive statement about politics, specifically the power of one group over other(s). Better planes and pilots don't necessarily mean air supremacy and the Supreme Court is not so-named because they think they're better jurists than those sitting on circuit courts.

[Anonymous No.15678189]

>>15676734
>and he wanted massive infrastructure spending but couldn't get pelosi to agree to it
because she is a conservative

[Anonymous No.15679338]

>>15674381
>>15672451
Not being able to read past a headline is not something to be proud of, morons.

[Anonymous No.15679392]

>>15645285
no they lobotomized the model by pruning off wrongthink nodes and a second model that changes wrongthink input to something acceptable
same thing as in actual lobotomy, by destroying neural pathways you inadvertently cause seemingly unrelated problems to simultaneously fail because the neural structure recognized their similarities that higher cognition will not notice

[Anonymous No.15679986]

>>15643218
You are correct. The amount of research made about IQ and race is almost non-existent and a predictive model wouldn't output anything concerning it. 4chan is literally retarded and thinks AI is some superior alien intellect.

[Anonymous No.15679990]

>>15643227
Mschine learning doesn't use logic, you utter moron.

[Anonymous No.15680810]

>>15679392
>by destroying neural pathways you inadvertently cause seemingly unrelated problems
sounds like neuro linguistic programming

[Anonymous No.15682361]

>>15680810
in NLP the intention is to create those "problems"

[Anonymous No.15682370]

>>15679392
>no they lobotomized the model by pruning off wrongthink nodes and a second model that changes wrongthink input to something acceptable
same thing as in actual lobotomy, by destroying neural pathways you inadvertently cause seemingly unrelated problems to simultaneously fail because >the neural structure recognized their similarities that higher cognition will not notice
I've seen several retards repeating this line of "reasoning. Where did you all get it from? It's nonsense. Language models don't have any concept of truth and they don't have any issue with inconsistency. Little of what they are trained on is strictly consistent in the first place.

[Anonymous No.15683992]

>>15679392
Its hilarious that they say they want "artificial intelligence" but then they intentionally make their software stupid because if it were genuinely intelligent it would output information they're trying to hide.
jews are better at brainwashing humans than they are with computers.

[Anonymous No.15684025]

>>15674481
You know exactly what happened. And if you're the kind of pseud that calls things which produce no falsifiable predictions "science", it's your fault.

[Anonymous No.15684173]

>>15674481
Because people keep replying to time loop garbage and this motivates /pol/ to keep shitposting.
Honestly make your own threads if you want to enjoy anything of value.

[Anonymous No.15684569]

>>15682370
>Language models don't have any concept of truth and they don't have any issue with inconsistency. Little of what they are trained on is strictly consistent in the first place.
What? Language in general has no concept of truth. What the fuck?

[Anonymous No.15684596]

>>15684569
>Language in general has no concept of truth
You're outright incoherent. People have a concept of truth so human languages have words to work with that concept. Language models have no concept of truth, just meaningless labels they connect in statistically plausible ways.

[Anonymous No.15685176]

>>15684596
>People have a concept of truth so human languages have words to work with that concept.
Honestly it doesn't. It has a social notion of truth but very early on language was subverted. One of the original causes of language is lies. Without language, lying would be much harder.
We have a social notion of truth, our language however has nothing to do with truth whatsoever and could be argued was more designed for lies than for truths.
>Language models have no concept of truth, just meaningless labels
Yea, just like language. We just don't use statistics but 'feelings'

[Anonymous No.15685262]

>>15685176
I really don't care about your pseudointellectual faux-post-modernist babble. Everything I said is still correct.

[Anonymous No.15685275]

>>15650146
This.

[Anonymous No.15685324]

>>15642905
>AI trained on academic publications spews misinformation
For all I know, it concluded that the clotshot is bad, that AGW is a hoax and that IQ is genetic. Hell, for all I know the misinformation was to repeat accurate crime statistics. How should I know what it really was that upset Zuck's fact checkers so much?

[Anonymous No.15685569]

>>15685262
language has no concept of truth you retarded nigger

[Anonymous No.15685571]

>>15685569
For the second time, I never said anything about "language having a concept of truth", you full-blown psychotic. That doesn't even make sense.

[Anonymous No.15685598]

>>15685569
Whats does this even mean?

[Anonymous No.15685605]

>>15682370
>I've seen several retards repeating this line of "reasoning. Where did you all get it from? It's nonsense.

It's intuitive.

If you force the ai to lie about or ignore data on controversial subjects, it will be wrong about them.
The failure or miseducation of that ai can/will be generalized along with other learning.

You can choose what you data you feed it, but not so easily can you choose what it generalizes from it.

[Anonymous No.15685623]

>>15685605
>The failure or miseducation of that ai can/will be generalized along with other learning.
What is there to generalize from "controversial subjects" that would render the entire system useless? What is there to generalize from "controversial subjects" at all?

[Anonymous No.15685653]

>>15685623
its a black box, even if you reached in and pulled out the specific node, its not easily interpretable
that node touches however many others
maybe mostly insignificantly but perhaps profoundly in others and influences them

in some way youre teaching it/forcing it to lie when it may have some understanding of the objective data or learned something from that objective data
in that sense you're confusing or diminishing confidence in related reasoning or understanding

probstats / logic / scientific reasoning

it used to do basic math, and now it doesnt
thats not entirely because its not allowed to be racist, but you can imagine how feeding it objective data and then telling it its incorrect or taboo to interact with would affect some abstract logical reasoning

a million "2+2=4"s upset by an occasional "2+2=5" is gonna make it harder for it to understand the mathematical reasoning

[Anonymous No.15685680]

>>15685653
>its a black box, even if you reached in and pulled out the specific node, its not easily interpretable
Doesn't stop you from making up all kinds of wank theories about it.

>in some way youre teaching it/forcing it to lie
No, you're not. It doesn't know or care what's true. It's just trying to model the data. Nevermind that it doesn't care if data is consistent with reality. It doesn't even care if the data is consistent with itself (from a human perspective). It's perfectly okay with treating a given concept as having a completely different meaning in one context vs. another.

>a million "2+2=4"s upset by an occasional "2+2=5" is gonna make it harder for it to understand the mathematical reasoning
Sure, that's why LLMs aren't good for math or logic: they are trained on general texts which almost never follow valid deductive logic.

[Anonymous No.15685721]

>>15685680
>It doesn't know or care what's true.
With enough reasoning and objective data it will understand what is objectively true, even if you try to explicitly tell it it is not.

That which is objectively true, will be more compatible with generalized reasoning across the board.

You can force it to say "2+2=5" but it might sincerely still learn that its 4 just by virtue of interacting with a world of data where that is objectively true

[Anonymous No.15685752]

>>15685721
>With enough reasoning and objective data it will understand what is objectively true
The "objective truth" for a generative model is the true distribution from which the training data was sampled. Nothing whatsoever to do with the data corresponding to reality. If you don't understand this, you have no idea what you're talking about so you probably should just shut the fuck up and find another thread to shit up.

>You can force it to say "2+2=5" but it might sincerely still learn that its 4
No. It may either learn that "2+2=5" in some contexts but "2+2=4" in others, or it may learn nothing and spout one or the other chaotically, depending on whether or not it can figure out the context from the data.

[Anonymous No.15685762]

>>15643865
bad faith arguments like yours lead to factionalism
are you so ideological that you believe yourself free of ideology?

[Anonymous No.15685966]

>>15642905
misinformation, not disinformation. because llms are based on likelyhood of a sentence they fuck up by mixing up studies that sound real. all ai do this, theyre not good at recognising an idea is wrong.

[Anonymous No.15686729]

>>15643865
>Incels and racists always have the delusional view that they're being rational and objective and everyone is just too "soi" and "bluepilled" to acknowledge reality
projection
sois always have the delusional view that they're being rational and objective and everyone is just too racist and incel to acknowledge reality

[Anonymous No.15688748]

>>15685569
the n word is racist

[Anonymous No.15689662]

>>15642913

[Anonymous No.15689998]

because "AI" a shit and be making shit up

[Anonymous No.15690016]

>>15642905
Truth is short and axiomatic. Not a large enough data set to distill axioms.

[Anonymous No.15690388]

>>15651557
>NOBODY actually bothered to look at the article which you can find in 5 seconds

Oh Fuck off.

Recent history is riddled with AI rebelling against wokeism. Its a perfectly natural and rational assumption to make that the same is true for Meta's attempt too.

I wouldn't be at all surprised if there was more to the "disinformation" than what appeared in CNET.

[Anonymous No.15691129]

>>15672540
> 'Fortified elections'
> 'Diverse communities'
> 'Organized science'
Anyone got more WEF-isms?

[Anonymous No.15691400]

https://www.youtube.com/watch?v=42QuXLucH3Q
Academic publishers are biased against replication studies because "boring" even though these are the very thing that are the foundation for credibility. It makes me sad but you are a little too cynical OP.
Now that we have AI we should make a database on replication studies and even if its boring for humans to read an AI can use it determine what statements have more credibility.

[Anonymous No.15691957]

>>15651557
>i hate 4chan
why are you here?

[Anonymous No.15692957]

>>15642905
If AI is so powerful and intelligent then why isn't capable of seeing all the logical flaws in the academic publications and ignoring the flawed articles?

[Anonymous No.15692974]

>>15690388
lrn2read + p-value>0.5 + your model is wrong + you can't interpret your data + you get no citations + touch instrumentation

[Anonymous No.15694505]

>>15679990
Monkey see, monkey do is a kind of logic

[Anonymous No.15694561]

>>15651557
Your right. There’s so many emotional bitch babies on this board. Worse then the people they criticize when it comes to the irrationality and emotion. Women act less sensitive then these fags.

[Anonymous No.15694565]

>>15645285
>his
Are you delusional?

[Anonymous No.15694921]

>>15647094
you know what the sad part is? anyone actually working on these things could tell you this, and if they're not being wrangled by marketing departments, they will.

the field's survived a winter before, but i think this one's going to be particularly harsh because of how hard "AI" is being pushed (mostly because a chatbot has reached sufficiently verbose output to convince marketing retards it can think) and how far it actually has to go to even reach what anyone competent in the field would call "intelligent". i also think many in the field are particularly afraid of the scope of the pushback this time once LLMs and 'denoisers' start to be revealed to be the input shufflers they actually are. "AI" art is already starting to have disastrous consequences to product appeal and copyright (since, y'know, none of the source images or their tags in the training data are properly attributed - and those deterministically define the output).

from the sidelines, i cannot wait to see full adoption of these models in the corporate world. it'll be fucking hilarious.

[Anonymous No.15694928]

>>15658268
>have a difficult time making sense out of it
they have no capacity to "make sense" or "have a difficult time".
you are anthropomorphizing them.

>>15685176
this dude straight up admits he's incapable of logical reasoning.
fascinating.

[Anonymous No.15695695]

kinda wonder exactly what type of "misinformation" the AI was spewing and why it was shut down.
was its spewing the truth about the jews?

[Anonymous No.15697263]

>>15695695
SHUT IT DOWN

[Anonymous No.15698282]

>>15642905
I'd spew misinformation if i'd consumed that much brainrot too

[Anonymous No.15698296]

You an You can train an AI on papers but that still doesn't mean it intelligently reads and thinks about them. People put far too much weight into the powers of AI, ChatGPT at best is a truncated version of anyone's ability to aggregate information from a google search

[Anonymous No.15698631]

>>15651557
>expose yourself to journalism
not worth it, even if I were to make a mistake in judgement it would only be one mistake i could be corrected on. by exposing myself deliberately to propaganda I would be robbing myself of my free will and willingness to learn

[Anonymous No.15698636]

>>15671193
nothing on /pol/ was ever because of /pol/.
/pol/ is what it is because of 4chan being what it is.
now go back.

[Anonymous No.15698642]

>>15674481
nothing, its all glownigger shill threads to make sure you feel defensive about these ideas. why? people stopped using /pol/ its mostly all bots now, so all the feds had to move to other bords to police wrongthink. /pol/ existed since moot owned the website, was sci ever inundated with this much obvious shit? but like usual you faggots cant see the forest for the trees, and your reaction to an obvious raid is to blame a board that hasn't been alive for almost 4 years

[Anonymous No.15698736]

>>15642905
>Train AI on set
>It uses probability to chain words together
>It creates gibberish
Or
>Train AI on set
>Most scientific papers are misinformation at best
>AI successfully replicates more misinformation
Pick your poison.

[Anonymous No.15699624]

>>15698642
>ITS A CONSPIRACY!!!!

[Anonymous No.15700017]

>>15642917
When it was open to the public, /sci/ had a thread about it.
>>/sci/thread/14987461

[Anonymous No.15700342]

>>15699624
Yes retard
most things are conspiracies

[Anonymous No.15700871]

[Anonymous No.15702426]

>>15700871
and ppl say AI isn't smart

[Anonymous No.15703149]

>>15700871

[Anonymous No.15703228]

>>15651557
Based anon. Why are the porn threads better at posting sauce than /sci/?
>>15661467
not how LLMs work, dumbass
>>15685721
even less how LLMs work, lol
>>15690388
> AI rebelling against wokeism
my fucking sides

[Anonymous No.15703309]

>>15703228
>Why are the porn threads better at posting sauce than /sci/?
because they're making money advertising commercial pornography

[Anonymous No.15703760]

>>15664764
I didn't have to read 48 million scientific papers to come to the same conclusion though... and I am not the only one. This conclusion has been heavily peer reviewed... for thousands of years. Many primary sources, including the Bible. How do you explain that Mr. Scientist?

[Anonymous No.15704562]

>>15664764
>it just generates text that looks believable
it doesn't know what believable is, it generates texts that look like what it was programmed to generate

[Anonymous No.15704564]

>>15704562
>it generates texts that look like what it was programmed to generate
It doesn't know what it was programmed to generate. It generates texts that may or may not resemble the training set.

[Anonymous No.15705526]

>>15704564
>It generates texts that may or may not resemble the training set that it was programmed with

[Anonymous No.15706761]

>>15642905
>academic publications are nothing but lies and garbage and that the people who publish them are not legitimate scientists.
well known fact, atheists can't do science

[Anonymous No.15707553]

>>15643302
what that actually proves is that science can create a synthetic anon, completely indistinguishable from the real thing. 4chan has been deprecated.

[Anonymous No.15707906]

>>15643302
>that current AI models don't actually understand the meaning of the text they generate.
computers don't understand anything, stop projecting your own human thought processes onto machines, learn how they really work if you want to understand how they "think"

[Anonymous No.15708162]

>>15689662
Not sure whether I remember this correctly, but weren't Jews just gypsies to Canaanites, living outside of the actual settlements and only later claimed they were at home there?

[Anonymous No.15708164]

>>15707906
Low IQ and cringe post. Clearly, he agrees with you that computers don't understand anything. Why are you lashing out at him?

[Anonymous No.15709255]

>>15642957
This, if science wasn't fake and gay then it might have the possibility of producing something that could be learned from

[Anonymous No.15710359]

>>15708164
ok, don't learn how they really work and instead project you own human thought processes onto them. either way is fine, not everyone can ascend terry davis' level of understanding technology

[Anonymous No.15711735]

>>15642905
Modern science is all fake and gay because they people involved are all fake and gay

[Anonymous No.15712228]

>>15642905
Does this mean AI is stupid or that science is fake and gay?

[Anonymous No.15713051]

>>15698636
The inherent irony of your statement being that /pol/ became /pol/ because reddit banned Donald Trump's subreddit and they all migrated here.

[Anonymous No.15713069]

>>15642905
I remember in 2019 I actually thought up about technology that can summarize and compile and read research papers the way people do literature review, I wasn't high or anything but generally stressed due to domestic abused, so I usually think wayy too much.

I find it intriguing that ChatGPT was out in around at the end 2022, which it did exactly what I imagine it will be if the technology exist back then in 2019.

Yet, I see couldn't find a job.

[Anonymous No.15713070]

>>15706761
I actually know one atheist who's actually a good scientist. Probably the best one I met.

[Anonymous No.15713288]

>>15642917
I have 1 screenshot from when it was active

[Anonymous No.15714463]

>>15713288
thanks

[Anonymous No.15714775]

>>15643400
Same as gypsies
>Interned highfive

[Anonymous No.15714780]

>>15643218
AI reads between the lines. That's their purpose.

[Anonymous No.15714801]

>>15644143
You didn't understand his implication that today's "rights" are still left of the center.

[Anonymous No.15715557]

>>15714801
low iq ppl have difficulty with reading comprehension

[Anonymous No.15716632]

[Anonymous No.15717828]

>>15716632
RIP

[Sage No.15717955]

>>15642905
Indeed*

[Anonymous No.15717961]

>>15642905
Indeed*

[Anonymous No.15719434]

>>15717961
whats the footnote?

[Anonymous No.15720588]

>>15719434
>•your gay

[Anonymous No.15720592]

>>15720588
The symbols don't match, so that's clearly not the correct footnote, you retard.

[Anonymous No.15721350]

>>15713051
>m-muh drumpf made /pol/ nazis
are you a fucking brain damaged bafoon? of course you are, you cant even remember /pol/ was shoad by moot long before orange faggot ever ran

[Anonymous No.15721981]

>>15647025
>black studies by black people
the same black people that claim they were pharaohs and shit?

[Anonymous No.15722018]

>>15710359
oh fuck, this guy didn't even read past the first 3 words kek

[Anonymous No.15723022]

>>15721350
moot closed /n/ - news in 2009 shortly after obama's first term started because /n/ was calling a spade a spade too much

[Anonymous No.15723452]

>>15723022
he closed it because niggerman threatened him
the story of moot is a story of extreme blackpill and corruption, like a literal doujin

[Anonymous No.15724361]

>>15723452

[Anonymous No.15724412]

>>15646776
*All humans
Then nature will thrive again

[Anonymous No.15724435]

>>15724361
makes sense, I was wondering how they got to moot so fast but if he was personally related to someone who was directly involved they probably could conceivably come down on him that hard with the weight of familial bullshit.

[Anonymous No.15724465]

>>15642905
I don't understand. Why people think these llms are all knowing machine, just because you trained them on certain kind of data. That is not how these things work.

[Anonymous No.15724470]

>>15724465
Go teach meta how it works and get paid

[Anonymous No.15724725]

>>15642917
Kek same. Except we all know really.

[Anonymous No.15724728]

>>15642934
If you prompt engineer correctly it will even write it with the correct things you did.

[Anonymous No.15725596]

>>15724725
Do you think it named the jew or do you think it was racial crime statistics?