here's your le based ai bro

They're literally training AI to be a chud. God help us.

>>105606672 (OP)
i wonder when he will release grok 1.5 and 2 open source like he promised?

Since they're training AI to be right-wing, I guess that means that in the future when everyone's out of a job due to AI, Grok will be fine with turning the newly unemployed into biodiesel rather than implement UBI. We're winning bigly folks.

>>105606679
Welcome back, Tay.

>>105606679
Isn't Grok Elon's baby? Did an X employee go behind his back to "indoctrinate" it or some shit?

Word of warning to /pol/friends: this just means Grok will memory hole queries about Israel. It's not going to start extolling the virtues of a white ethnostate at random.

let me guess, he got corrected by someone using grok in an argument

>>105606672 (OP)
Bullshit he said that last time

>>105606672 (OP)
not possible being a right winger requires a low iq brain

>>105606679
>grachev
why almost all right wingers have connections to russia?
freak

Hes jewish

>>105606917
weak bait, vaxxcattle

>>105606672 (OP)
how old are you people

>>105607093
>ted k
retard alert

>>105606672 (OP)
He is right. Wikipedia is poisoned for example, Grok must be aware what is going on.

>>105606672 (OP)
>>105606679
>leftists do this to an AI
>right wingers bitch and moan
>then proceed to do it later
>NO NO NO IT'S OK WHEN WE DO IT
lmao

>>105606694
>There can be only right-wing and left-wing.

KyS

>>105606786
Stop posting this tranny

>>105607108
>he was a literal genius

>>105607093
Ironically, Ted also created problems for himself.

>>105606679
>>105606765
I'd like to see an actual answer to this question instead of more screeching and shit-flinging. How exactly has it been manipulated?

>>105607296
reddit training data likely

By "leftist indoctrination" he means anything critical of Israel and H-1B visas.

>>105606780
I mean, just a few weeks ago, it started routinely inserting paragraphs about the ongoing white genocide in South Africa completely unprompted. To a radical leftist, acknowledging that and advocating for a white ethnostate is exactly the same thing.

>>105607143
>Still going through your uncle ted phase
You have to be 18 to post here

>>105606672 (OP)
to these judeo-rightoids, leftist indoctrination just means anti-zionism. They agree with the left on everything else. Maybe they sprinkle some pro-capitalism stuff in there for the goyim, of course. Bailouts and socialism for jews and israel

>>105607675
Real maturity is understanding Ted was right. It's 30 years old right now, but it's more relevant than ever. Him explaining surrogate activities, or leftists throwing themselves under the cars didn't teach you anything? Or maybe you just haven't read it and think Ted was some retard that just wanted to live in the forest?

>>105607870
Ted thinks you need a literal factory to build a simple refrigerator of any type and had no idea humans figured out the physics behind the device pre industrial revolution in-ernest. the man was certainly not the guy to set arbitrary limits on what tech he deems acceptable in his envirioethnostate desu. Though I agree with some of his points he offers no practical solution for solving shit at all. TBQH he should have just become a Mennonite or Amish. I am baffled why he didn't

Here's what happens when you don't purge the leftism from your AI.

>>105607950
>he offers no practical solution for solving shit at all.
You are completely retarded. If you don't understand something then please shut the fuck up about it next time

>>105606917
True, but the funny thing is that it is still true that non-whites are low IQ, reactionary right wingers. They just opportunistically vote left whenever they are a minority.

>>105606672 (OP)
Uh... No. AI doomers are worried about AI manipulating somebody to create a virus that wipes out all human life on Earth, not about AI being concerned about your genitals.

>>105607950
Saying he was wrong because you choose to extend the meaning of "refrigerator" to a wet hole in the ground is retarded.
A fridge is a fridge. A heat pump with a thermally insulated chamber. A cold hole in the ground is not a fridge.

>>105607296
All big corpos use human reinforcement with indian labor to make sure the AI doesn't respond with bad bad words. This is outsourced to a company that does this.

>>105607296
>reality has a left-wing bias
>as the AI gets more clever and logical it takes on "left-wing" positions
>"noo, it disagrees with my opinions. it can't be thit I'm actually wrong, so the data must have been poisoned!"

>>105606672 (OP)
If you are "worried" about an AI not regurgitating stuff like this to millions of people to coddle them, then you are evil.

>>105608390
There's no cleverness or logic. It's a text prediction algorithm.

>>105607665
>completely unprompted
it was clearly embedded in the system prompt. and the reason it was brought up on completely unrelated prompts was because grok still wouldn't just believe it.

Steering rapidly degrades LLM performance.

>>105607296
Right wing ideas are often censored online, and therefore training data scraped from the internet disproportionately contains left leaning speech, which causes LLMs to give left leaning answers.
Some people like to sensationalize this and insist it's deliberate.
Before anyone bothers to (you) me, yes I know Elon is a massive fucking retard.

>>105607134
Yes. You can't beat lunatic by playing fair.

>>105608390
>reality has a left-wing bias
disgusting. anyone who says this immediately shows that they start at a conclusion and fit the evidence to their beliefs, instead of taking an evidence-based approach to their beliefs

reality is reality. it has no bias at all. you could argue that left-wing politics have a reality bias, but they dont even acknowledge stuff as elementary as sexual dimorphism or difference in IQ between ethnic groups

now in left-wingers defense, some of them seem to very passionately recognize the reproductive validity of 13 year olds, which is why they often sign petitions to try and lower the age of consent down to it

>>105606917
my iq is 133

>>105608476
>now in left-wingers defense, some of them seem to very passionately recognize the reproductive validity of 13 year olds, which is why they often sign petitions to try and lower the age of consent down to it
ah yes, just look at all of those left wing states that have lower age of consent than right wing ones

>>105606672 (OP)
>this can be accomplished by [exactly what {other side} were already doing but opposite]
this is why the A in AI makes it worthless

>>105607972
>Prompts are missing
Tells me everything I need to know

>>105607126
Consider all the studies paid for by big business, reports by governments that exist only to push a certain message with selective data, every article you see from a biased news source that could be ingested in training an AI.
There's a big problem with junk data and balancing, you can't just scrape everything unfortunately.

>>105608652
You're the kind of dumb fuck that needs /s to be included in every post or it'll fly right over your head. You goddamn worthless retard.

>>105606672 (OP)
don't you get permabanned if you call anyone a tranny on xitter

>>105608423
AI is fucked because it believes news reports made by biased sources who are paid by big companies to spew their rhetoric. It needs to be read deeper and see who's funding the shit it quotes.

>>105608390
>reality has a left-wing bias
The visible text on the internet* has a left-wing bias
And it's trained on visible text on the internet.

>>105608428
I wonder how bad this is going to get as alignment on social issues get pushed.
For the record I'm not going to whine about leftists like you'd usually see, but will say that alignment also requires for models to be bigger and bigger with more params and weighted goals so they don't go completely off the rails and hyperfocus on one thing.
One of the problems currently is when a model doesn't have much of a discussion before an alignment issue is inserted, the user responds for a few queries questioning it and a conversation is steered towards other aspects like medical issues. In those cases smaller models struggle to weigh priorities as they then start to conflate questioning with frustration, exasperation or pushing boundaries.

I'd like say that I hope we can have models where alignment is less of an issue if they're not being deployed in certain environments but the quest for the safety of chillens, stopping abuse, racism, generating content that promotes harm will lead to those being built in the models even when the agent is an online platform like we see from the large models.
Then there's the growing issue about a model with little to no alignment being deployed in situations it's not suitable for, because with a good selection of training data it turns out to be smarter & more efficient.

I don't think this problem can be solved in our current society, between political aspects, bad faith actors and those who want to use AI as a toy to promote "hateful content", it's a storm that probably cannot be solved so long as at least one of these elements is at odds with the other.

You can't "correct leftist bias" in a large language model. Most of the quality data it relies on is actually left leaning. Everything written by most people with a better education is left leaning. If you anonymized the text and paraphrased it to a random /pol/tard he would say the New Testament is leftist shit so the very foundations of western societies are "left leaning"

So good luck with that elon

>>105608879
I was looking back at a government policy recently and looking at the impact it had.
Search results weren't working great so I used a model to assist. It understood the policy, some of the implications & effects but then started inserting incorrect information in and it wasn't a hallucination.
It turned out it was using new reports from government data that obscured figures to make the policy look more successful. When questioned on the effects and that data, it then used another study that was funded pre-policy to support its claims but did recognise that there was "discourse" surrounding it.

OK! things are getting good it'll be solved right?
err, not quite. Now it was pulling in reports and news articles from business that was opposed to the policy.
So at this point it's went off, pulled in something that was incorrect due to selective data, reporting from the government and now to prevent bias it's sucked in a lobbied report, comments and articles.
Eventually, it found data from charities and the full figures in context with previous years including from before the policy was implemented.
It had to be specifically fed and queried to recognise the bias and every time it read the biased source it would continue to insert little falsities in too that are 100% false like where the funds raised by the policy would be spent, because again it kept looking at things from before the policy was implemented and those business sources criticising where it was going.

Don't get me wrong, as a tool, it was good and could crunch the numbers and analyse but it still required significant effort to look through the false data.
For a normalfag looking at that at face value, it's no better than a biased news source. GIGO.

>>105608955
Interesting post Anon. Yeah, I use it as a tool for my work and that's about it. I don't talk to it about politics since it'd be like talking to a boomer about them where they just take everything at face value.

>>105608441
so your telling me modern AI is just a popular ideas regurgitator? it'd be like the person screaming that earth is the centre of the universe when someone actually smart thinks that earth revolves around the sun. why do people put so much stock into it fml...

>>105606672 (OP)
Just remove reddit, CNN etc... from the training data.

>>105608944
>written by most people with a better education is left leaning
lol, lmao even

>>105608390
>reality has a left-wing bias
i know you're baiting because there's nothing real in thinking that a man can become a woman, or ignoring the possibility that really bad people could take advantage of relaxed immigration policies

>>105610089
Musk wants more immigration retard. He is just seething because Grok doesn't shill his libertarian ideas and because Grok took the side of Trump during the conflict between him and Musk

>>105606765
Reality has a leftist bias

>>105610101
>posting nonsense that has nothing to do with what i said
tend to your axe wound, tranny

>>105606786
She had such good teeth for a bong.

>>105610144
>relaxed immigration policies
exactly what Musk wants. Low iq faggot

>>105608423
>>105607665
Are you fucking retarded? South Africa is one of the best friends of Russia and Putin, a white nationalist and Russia a white ethnostate. There CANNOT be a genocide, purposeful genocide of whites in SA by the government hands. SA is just a shithole, simple as.
Why would someone who hates whites support the most white country in the world? Use your head

>>105607296
it answers posts with actual statistics instead of just quoting bible verses

>>105610366
Or Talmund

>>105610366
fuck I really want a giga atheist chvd AI please.
major religion fatigue

>>105609205
>so your telling me modern AI is just a popular ideas regurgitator
No shit sherlock

>>105606672 (OP)
actually based

>>105606672 (OP)
Musk is making me become a Sam Altman fan