You'd think Elon would be familiar with the phrase "Garbage In, Garbage Out". Having a LLM decide what the correct version of the entirety of human knowledge is, deciding what's "wrong" and deleting "errors", in a recursive fashion, is a profoundly stupid idea, even for Elon.
Is it possible that Elon is not, and has never been, particularly smart? He seems to have stalled intellectually and emotionally at about 14.
Chronic drug abuse will do that.
I'm saying he didn't have that many neurons to sacrifice to ketamine (etc, etc, etc) to begin with. Drugs def did not help.
Having all that money and power has become meaningless to him. He wants to control all thoughts and history. This man is truly broken.
Also makes clear that he doesn't understand what an LLM actually *does* 😩
Please tell me that this is what Elon thinks is a funny joke.
Dude makes no sense
Retraining on the database it creates is another rung on the steps to the level of stupidity on display.
The fact that they even consider retraining on the data they gathered post-"corrections" implies they truly believe that knowledge is cyclic— part of a conspiracy and pseudoarchaeologic theory of "first humans" and all that grifter nonsense. Correction without knowledge is just censorship.
You can learn a lot from garbage - but you have to know its garbage.
Deep reflection and impulse control are not his strong sides.
He really needs to fade from existence. The sooner the better.
He's just trying to rewrite history by slapping a veneer of tech over it in an attempt to fake legitimacy.
Especially when Elon doesn't like when the LLM latches onto a truth he finds inconvenient.
Elon‘s biggest achievement is to convince billions of people he‘s really clever.
There seems to be a paradox here: What will be the source of the “missing information“ added to “the entire corpus of human knowledge“?
What I was questioning….
Translated: replace the truth with propaganda. This is why they are attacking Wikipedia!
AI will just further “dumb down” the already dumb.
What was he expecting from a LLM trained on Twitter?
If AI models train on their own slop does that make them worse than useless? bsky.app/profile/john...
yes, it causes model collapse; it's the habsburg chin of machine learning
Ah well, at least we now have confirmation that Elon doesn’t have so much as a basic understanding of what LLMs are and how they work.
Absolutely. One man's garbage...
Well he is a expert on producing garbage.
Like the dodge morons did with gov docs, they’ll probably try removing anything that says trans from the training data. Grok will end up insisting car transmissions don’t exist.
Elon Musk is a solid, weapons grade, profoundly stupid cunt. Anybody who worships him or buys shares in any company in which he is affiliated is a sad loser
Will we have perfect explosions on the launchpad?
And why delete the errors? An LLM would be perfectly capable of tagging information as trustworthy (validated) or not (who makes that decision is another matter). This looks more about hiding stuff Elon doesnt agree with.
I belive we call these kind of people revisionist - at least when it comes to holocaust etc. We tend not to like revisionist.
Musk is profoundly dangerous
it will produce worse output, but I do think folks should be super clear here too: what he is talking about is explicitly creating a propaganda engine
fact. (also, i remember that episode of babylon 5; too bad grok isn't as smart/independent as historical babylon 5 crew holos :p )
An alternate reality presented as fact, very Orwellian. We all hope future generations will learn from past mistakes, does Elon want to destroy that hope? Won't be long before philosophers & free-thinkers are rounded up.
In other words the history of the world according to Elon Musk.
I doubt there is any actual reasoning and the “advanced reasoning” is just a marketing tagline. They’re just going to code Grok to ignore information from sites they don’t like.
machinelearning.apple.com/research/ill...
It doesn't really matter tho. Simple example is suppose you have 100 billion documents that you use for training. Have the LLM go through them and label the 150 million that mention the phrase you don't like. That's a thing most LLMs can do relatively well without a ton of errors.
I suspect he’s going to pull his “white genocide” trick again and clumsily shoehorn in data that supports his preferences.
Then retrain on the 99.85 billion documents you have left over, and you've created an LLM that is unaware of those labelled 150 million documents.
What about using LLM to not skip undesired documents, but rewrite them, so the new model thinks they say something else?
You think they’re likely to do better than canceling research on “equity” and discovering they’ve bankrupted all business school finance departments?
A+ post
You've created an LLM there that is /worse/ because it is (a) smaller, and (b) intentionally skewed, but it's also not so much smaller that it will be visibly useless, and it has knowledge of other topics, even if it is much dumber about things adjacent to the filtered topics.
And then Elon will brand it as more “intelligent” and lots of people will believe it.
It doesn't require any particularly sophisticated steps to do it, or need an LLM that is particularly intelligent to help you get there, which is also partly what makes this dangerous: if you have enough money and an AI machine, you can absolutely construct propaganda models. And he's saying he will
I think you’re right, making propaganda models can be done. It already works that way of you know how to put the right things surrounding the LLM, though. All of the bad stuff is in there you can get it out with the right input prompts. This also shows Musk doesn’t know how they work.
Right, we've already seen several cases of people trying to create propaganda models in a few different ways. Musk did it already with system prompt hacking (the famous "grok is obsessed with white genocide" fiasco). DeepSeek did it too, but in post-training, objecting to Tiananmen square references
You have some filtering which is more neutral; prior to the 2024 election several online models tried to detect and bail when talking about politics to both avoid brand risks and "election disinformation", and genImages try hard to detect attempts to create abuse material
Which all gets back to the hostility Musk et al have for Wikipedia -- for its faults, Wiki is a largely reliable source of verified facts and data that resists being used as a propaganda machine.
quite an expensive fools errand too, training ain't cheap
but this is much more direct attempt to create propaganda: that the output of the model should seek to achieve political persuasion outcomes and be deployed to a site that hundreds of millions of people, including political elites, journalists, and citizens
I am certain Grok was already a propaganda machine before the election.
Musk should take a ride in one of his rockets.
An LLM is actually a perfect propaganda machine, it's real power is not the veracity of its output, but the convincing way it presents it. The training data is still the costly time consuming point I would argue, consensus on what it should spit out, and internal integrity of the data presented
The real power is that most users believe that it possesses intelligence. Calling it AI is a misnomer that needs to stop. It’s a really good mimicry machine.
They believe it possesses intelligence exactly because it's output sounds so convincing. it's output sounds like human speach, and with authority.
It's likely the same "reasoning" that other models are doing, where it outputs repeated text as if it is planning a response, and then responds. But yes, you're right that none of this is actually reasoning.
I love the Apple paper where they gave "reasoning" models problems that first year CS students solve with 5 line programs and the reasoning models all fall flat when the problem gets too big to regurgitate.
This approach was obviously going to be brittle and limited. I find it embarrassing that there are CS people ignoring decades of theory when considering these models. They act like we’ve discovered magic and not a probabilistic finite state machine.
His is more explicit, but in a sense they all kinda are. Lot of decisions being made behind the scenes influenced by bias no matter how hard the teams think they’re trying not to be.
This is not a defense of Elon, just my general problem with LLMs’ role in our disordered information ecosystem.
📌
He’s creating his own reality and wants everybody to accept his reality as THE reality!
tbf, it's not going to work
Perhaps, but it is still important that this is a thing he very sincerely intends to try and then to deploy on the site most politicians and journalists use for getting their information day-to-day
Also, even if it just works halfway, the target audience really wants it to work, they want it to spit out exactly what they want to hear. It's a self reinforcing mechanism, and those politicians and journalists are undoubtedly going to regurgitate at least some of the propaganda
concerning is his deleting of information his machine determines is "wrong" esp since he has scraped all of our information & all gov't databases
I'm surprised they didn't already curate the training material to cater to Elon's ideologies.
Exactly. There’s no truth, no facts. Reality is what he says it is.
Fox News for AI
Exactly, it's clear that he will be the one to decide what is wrong data and what should be added
He already has created an explicit propaganda engine
My worry would be that others follow suit or that it’s rep doesn’t drop (as MAGA leaners will already support something despite all facts) so it stays in the running and its garage then becomes part of the cannon of training data or bench marking slowly losing all validity of “known knowledge” 🙁
A Smithers that burns the planet.
Elon is straight up confessing.
It used to be known as “Garbage in, Garbage out” but in 2025 all the AIs started referring to it as “ole Muskie in, ole Muskie out” We have no idea why. 🥲
Musk isn't a genius. He's a drug addicted nepo-baby with an emerald mine, who buys geniuses, and ruins companies and lives.
Whenever he thinks he is being clever, Errol tells on himself. It takes a monster to create a monster.
See: Fred and Donald Trump
it's like trying to rewrite reality and correct history to suit a certain mind frame. Fox News already does that.
"Adding missing information." 🤔
What Musk is really saying is that he wants the ability to re-write human history in the image of the Third Reich!
He will make AI useless.
Omfg
He doesn't want Grok to be accurate. He wants it to spout garbage.
He is the equivalent of someone who thinks code is infallible because it passes a couple of unit tests
Seems to be profoundly 1984 to me. He's describing Winston Smith's job.
My thoughts exactly. ‘Retrain on that’ is sinister. I suppose he means something like retrain LLM systems but it has real forced re-education camps vibes.
“Then retrain on that”?
And "adding missing information" from -- someplace it can't access? 😕
so many recursive errors in logic there it makes one's brain cramp
"Adding missing information" to "the entire corpus of human knowledge". In other words, it's going to just make shit up, and then hallucinate on that, instead...
Whatever, the most important thing is that he will not be listed as a top-3 misinformation spreader on Twitter by grok anymore
Uncorrected data = inconvenient facts
Every so often Elon puts out a statement about a technical subject area that causes people with actual domain expertise to shake their heads in disbelief at how moronic it is and how stupendously ignorant of actual science and engineering he is. This is such a statement for AI researchers.
I asked five different AI models to interpret the Method of Everything manuscript - doi.org/10.3389/frma... - which reveals that without motion-> No potential-> No selection-> No intelligence. Do you know of anyone else that has done this?
Grifters Razor: never attribute to the stupidity of the poster what is actually the stupidity of the intended audience.
Musk wont stop until the robot agrees with him about everything, because only he has the knowledge and understanding of the world, no one else.
“Sir, it’s still not saying that you’re cool.”
I know what you're saying, and am not criticizing --agreeing, rather -- but why would anyone think that he's know that? he's proven *repeatedly* that he knows almost nothing at all and has no desire to change that
Also, where is the „missing information“ supposed to come from, if not from the „entire corpus of human knowledge“? That just translates to „We‘ll just make shit up“
The Stalin version of Grok.
This is key history is about evidence and interpretation of meaning. AI is not that good at it.
"The entire corpus of misinformation I favor" -Elon Musk
More like "Drugs in, Garbage out"
NINO nazi in, nazi out.
He's a drug addicted maniac.
But what Musk's actually saying is not that Grok will make those decisions, Musk will, and that hellscape will be Grok's new source of truth. his actions have basically telegraphed this sort of thing for a good while now.
No, it’s simply fascist.
He just wants Grok to be viewed as an objective oracle and for that oracle to tell humanity why combining AI and Eugenics is right and good. This unfortunately does work on too many already.
Did anyone ask Grok what could go wrong with that strategy?
Excuse me! Us stupids don’t want him either.
And half of that garbage probably comes from his own tweets….
Yet another Muskish disaster in the making.
I would expect that behind the scenes he’s going to make a series of inconsistent personal interventions reflecting his foibles, ideology and personal version of events ie the apartheid regime will seem less bad, Afrikaners will become permanent victims
And his rockets still keep blowing up too! 💥
seems like hes experiencing drungs induced psycosis. Not uncommon for people who abuse ketamine and other anethetics.
So basically: Grok keeps fact-checking him because it is trained on reality - let's rewrite reality.
This is not harmless stupid - it's dangerous stupid. Btw. have you heard of the white genocide in South Africa?
He's going to add the Protocols to training set isn't he
