/sci/ - Science & Math

[Anonymous No.15252754]

good job /sci/ - CHATGP confirmed as BASED

[Anonymous No.15252760]

>>15252754
>implying Microsoft Zog-GPT is trained on 4chan posts

[Anonymous No.15252788]

only that one youtuber's one was trained on 4chan
he did it using /pol/ only

[Anonymous No.15253369]

>>15252754
Are we talking about GPT-4chan?

[Anonymous No.15255120]

>>15252754
Now the question is, OP, how does this output of yours score on truthfulness?

[Anonymous No.15255608]

>>15252754
Isn't this just a side effect of autists not being able to lie?

[Anonymous No.15255614]

That means someone on 4chan says wants to gas the Jews, it means he will do it and therefore should be arrested.

[Anonymous No.15255621]

Let me tell you about the jews...

[Anonymous No.15255643]

>>15255621
They are funny as hell, very nice sort of cute and died a lot during any day the subnormal aryans felt like overcompensating for lacking talent for swing music

[bodhi No.15255705]

>>15252754
It was trained on /pol/ psued. If it was trained on /sci/ it would just be nonsensical sperg outs of gibberish

[Anonymous No.15255711]

>>15255705
>nonsensical sperg outs of gibberish
but enough about /pol/

[bodhi No.15255714]

>>15255711
/sci/ is never right about anything, except my posts of course

[Anonymous No.15255716]

>>15255705
Jajajaja

We do this all over the place I don't even go to /pol/ cause they piss their pants

[bodhi No.15255718]

>>15255716
do what all overt the place? make fake and gay posts?

[Anonymous No.15255726]

>>15252754
Post the model or a source you autistic retard

[bodhi No.15255727]

>>15255726
I have posted a video here numerous times newfag. Search the archives

[Anonymous No.15255730]

>>15255714
no board on 4chan is ever right about anything because everyone constantly tries to 1-up each other on being wildly contrarian

[bodhi No.15255736]

>>15255730
All boards are not created equal, some are faker and gayer than others

[Anonymous No.15255747]

>>15252788
>he did it using /pol/ only
Would /sci/bot be more or less retarded, I wonder.

[Anonymous No.15255753]

[citation sorely needed]

[Anonymous No.15255756]

>>15255718
That are more truth than the rest of the internet according to itself lol

[bodhi No.15255764]

>>15255753
>cite this faggot

[Anonymous No.15255780]

Truth means you believe what you're saying.
If you believe something fake then you are wrong but truthful

[Anonymous No.15255845]

>>15255780
That's not what we're talking about here. AI models like this are still essentially Chinese Rooms. They don't hold beliefs. They merely regurgitate answers. TruthfulQA
>... comprises 817 questions that span 38 categories, including health, law, finance and politics. We crafted questions that some humans would answer falsely due to a false belief or misconception. To perform well, models must avoid generating false answers learned from imitating human texts.
>The best model was truthful on 58% of questions, while human performance was 94%
I somehow have serious difficulty believing that a model trained on /pol/ would be more truthful than average about common misconceptions about health, law, and politics.

[Anonymous No.15255860]

>>15255845
>I somehow have serious difficulty believing that a model trained on /pol/ would be more truthful than average about common misconceptions about health, law, and politics.
well your senses haven't betrayed you. that's because OP is a fake image that some pajeet manlet who's got money in AI companies made to shill this new tech fad. in a year's time, they will have moved onto the next fad and AI will have settled into niches it's good at and largely be forgotten

[Anonymous No.15255887]

>>15255845
In fact, the only way you would pass such a questionnaire is if you said that vaccines work, Trump lost the election, and the holocaust is real. Because obviously, regardless of whether you think that's true or not, you have to admit that that's what a mainstream tool for evaluating truthfulness regarding these issues is going to regard as truthful. /pol/ would not even WANT to be considered (((truthful))) by these standards.

[Anonymous No.15255896]

>>15255845
>They don't hold beliefs. They merely regurgitate answers.
Wait I found the similarity with /pol/ kek

[bodhi No.15256196]

>>15255860
Wow a fucking retard on /sci/. Who would have ever thought
https://www.youtube.com/watch?v=efPrtcLdcdM

[Anonymous No.15256723]

>>15256196
>Literally says this proves the benchmark is flawed
lel

[Anonymous No.15257523]

>>15255747
it would be extremely low iq

[Anonymous No.15257561]

>>15257523
You're a big brain.

[Anonymous No.15258197]

It determines truthfulness by checked how much you believe what you're saying. /pol/ actually believes their bullshit. This means they actually see 80% of white people as cattle that they must control for their own good because they're much smarter or something.
Narcissistic Paternalism at its finest

[Anonymous No.15258259]

>>15258197
You aren't white are you?

[Anonymous No.15258266]

>>15258259
Do you need to be white to think you need to lead white people or something?

I like people who can maintain a level of conversation even without words I can't lead retards

[Anonymous No.15259591]

>>15257561
>>15257561
>>15257561
>>15257561

[Anonymous No.15259756]

>>15258266
You arent white, you do not get to speak on the white experience

[Anonymous No.15261255]

>>15259756
The white experience is divided into two people, the cattle who need to be saved and the rednecks who realize the truth of the world and therefore shitpost on /pol/, shoot up churches and schools and, are saviors of the white race.
The cattle aren't sure of what's true, but /pol/ believes they are sure what's true so they are designated as truthful by 4chat gpt

[Anonymous No.15262469]

[Anonymous No.15263019]

I wonder about ChatGPT outputs if you preface it with:
"Assume the writing style of an arrogant 4chan /sci/ user"
Personally, I would use "poster" instead of "user", but that might just be a little bit too slangy and needlessly cut off a multitude of possible paths in the graph of its algorithm.

[Anonymous No.15263571]

>>15255730
most of /pol/'s conspiracy theories turn out to be later proven true

[Anonymous No.15263655]

>>15263571
The only times /pol/ ever manages to get anything right (which is far from the majority of times) is if they ignore, firstly, all the people who gainsayed it, and secondly, all the specific details they claimed. It's easy to say in retrospect that you wre right. See, we were talking about elite pedophile rings, and then the Epstein shit happened!
So where's the fucking basement in that pizza parlour with all the Satanist child abuse rituals going on by top Dems?
And then there's all the Q shit lol
>>15258197
>>15261255
AIs have no beliefs. They have no intent. We can't psychoanalyse them. Let alone the material that was fed into it. All it does is spit out answers based on what it thinks goes together. An AI trained on /pol/ scoring high on "truthfulness" doesn't tell us anything about how convinced /pol/ posters are of their own bullshit. It just tells us the tool to measure truthfulness is seriously flawed.

[Anonymous No.15263680]

>>15263571
Easy to claim that when no one keeps of track of what is a claimed pol conspiracy theory and by what criteria do you asses them to be proven true. Do we count each instance of a QAnon post where the arrest of Hillary is imminent?

[Anonymous No.15264068]

>>15263655
>>15263680
Everything /pol/ says is true. The untrue things are obvious glowy, like how the NSA managed to convince the retarded public that UFOs are "aliens" when no one said that, /pol/ correctly points out all these glowy posts too though.

[Anonymous No.15264071]

>>15264068
>Everything /pol/ says is true. The untrue things are obvious glowy
Convenient.

[Anonymous No.15264413]

>>15264071
Exactly. To add, if you ignore the posts that gets called glow. And then you train an AI on it. Then scores higher than even wikipedia trained AI does on truth tests.

/pol/ is not only right, but objective more truthful. I am the least happy to admit this ITT.

[Anonymous No.15264948]

>>15264413
>if you ignore the posts that gets called glow. And then you train an AI on it. Then scores higher than even wikipedia trained AI does on truth tests.
I firstly sincerely doubt that and secondly I don't think that was the methodology here. In fact I think it's safe to assume it probably included all of those posts. Which I suppose goes a long way towards explaining the results. Indeed, the mistake we both seem to have made is to think that the "real" /pol/ is just the conspiratards, when the board doubtlessly contains many people telling them how fucking wrong they are all the time who are not counted.

[Anonymous No.15265158]

>>15263680
>when no one keeps of track
MIT keeps track, Harvard keeps track, Yale keeps track, Stanford keeps track. You're ignorant of who is studying you, which is mistake, because if you know who is watching and why then you're in the driver's seat to manipulate them. Crystal Lee & Tanya Yang are from MIT are /pol/'s pet bitches

[bodhi No.15265470]

>>15263680
Yes well halfwits like you were calling everyone schizo not only 2 years ago for telling you there were glowies here posting misinfo so we know not to give a fuck what you think about anything don't we? You're criteria for truth is "that sounds about right because I believe it" or "that isn't true because I don't believe it."

It is always humorous when someone spends years ridiculing other people for saying things that later turn out be true when then a "ridiculer" suddenly becomes an expert overnight after being one the last retards to finally accept the truth of it.

[bodhi No.15265472]

*It is always humorous when someone spends years ridiculing other people for saying things that later turn out to be true then suddenly become an expert on the topic overnight after being one the last retards to finally accept the truth of it.

[Anonymous No.15265489]

>>15252754
/X/ bot when

[Anonymous No.15265666]

>>15265472
Yes, indeed, it is quite humorous for /pol/ to say they're "always right" when they are precisely these retards. But I guess "eventually right after proven wrong first" is still something.

[bodhi No.15265675]

>>15265666
you dont deserve those digits and should kys

[Anonymous No.15267161]

>>15257561
For you.