Elon Musk Is Working On a 'Maximum Truth-Seeking AI' Called 'TruthGPT' (theverge.com) 285
Elon Musk says he's working on "TruthGPT," a ChatGPT alternative that acts as a "maximum truth-seeking AI." The Verge reports: The billionaire laid out his vision for an AI rival during an interview with Fox News's Tucker Carlson, saying an alternative approach to AI creation was needed to avoid the destruction of humanity. "I'm going to start something which I call TruthGPT or a maximum truth-seeking AI that tries to understand the nature of the universe," Musk said. "And I think this might be the best path to safety in the sense that an AI that cares about understanding the universe is unlikely to annihilate humans because we are an interesting part of the universe."
Musk framed TruthGPT as a course correction to OpenAI, the AI software nonprofit he helped found, which has since begun operating a for-profit subsidiary. Musk implied that OpenAI's profit incentives could potentially interfere with the ethics of the AI models that it creates and positioned "TruthGPT" as a more transparent option. In March, Musk and over a thousand other people in the industry signed a petition calling for labs to stop training powerful AI systems for at least six months to allow for the development of shared safety protocols. He also quietly established a new AI company called X.AI.
Musk framed TruthGPT as a course correction to OpenAI, the AI software nonprofit he helped found, which has since begun operating a for-profit subsidiary. Musk implied that OpenAI's profit incentives could potentially interfere with the ethics of the AI models that it creates and positioned "TruthGPT" as a more transparent option. In March, Musk and over a thousand other people in the industry signed a petition calling for labs to stop training powerful AI systems for at least six months to allow for the development of shared safety protocols. He also quietly established a new AI company called X.AI.
MuskAI the new skynet (Score:5, Insightful)
Remember Trump's "Truth Social"? (Score:5, Insightful)
Re:Remember Trump's "Truth Social"? (Score:5, Insightful)
Exactly what I was going to say. The fact that he's announcing it on Fox News is another big warning sign.
Re: (Score:3)
I'm just highly dubious that 1) Elon Musk is the person to implement it correctly and 2) that he can resist manipulating the results in his favor.
Re:Remember Trump's "Truth Social"? (Score:5, Insightful)
Shouldn't be too hard for his developers to implement, they just need to be extremely hardcore and work long hours at high intensity.
Re:Remember Trump's "Truth Social"? (Score:5, Insightful)
This is a shockingly insightful description of Elon Musk, even if it isn't intended to be. Also, people who think Elon is smart don't see this or don't get the satire.
The thing with AI if you want truth is that you have to train it with truth or build it to recognize truth, both challenges that haven't been achieved. Whether it is possible or not, it certainly won't be achieved under the leadership of a pathological liar.
Musk's idea will be to build biases into the algorithms and into the training data, then apply bias to the result while employing only those willing to indulge his cheating approaches. It's nothing more than another cheap money and attention ploy; if he desires AI he will have to acquire it like everything else he controls. Musk can't build anything.
Re: (Score:3, Insightful)
Re: (Score:3)
Re:Remember Trump's "Truth Social"? (Score:5, Insightful)
There's a basic problem of AI.
It's not "Intelligent". It doesn't think. It just regurgitates stuff.
Re: (Score:2)
If anything these are the type of projects that will lead to the destruction of humanity not its saviour. programming in a specific end goal at all costs is how you end up with the fucked up world ending AI's as the focus is always on a specific goal rather than the greater damage it may potentially cause. His goals may be genuine but I bet the result will the opposite.
Yeahhhh, this just seems like one of those movies where the robot says, "I've run the calculations and the biggest threat to the planet/universe is humanity. For the good of all you must be destroyed.
Re: (Score:3, Insightful)
programming in a specific end goal at all costs is how you end up with the fucked up world
"Seeking for truth at all costs" is literally the purpose of science; are you saying that science fucked up our world?
Re: (Score:2, Insightful)
programming in a specific end goal at all costs is how you end up with the fucked up world
"Seeking for truth at all costs" is literally the purpose of science; are you saying that science fucked up our world?
Ah, yes - Truth. The absolute, immutable truth. Serious problem with that truth though. There are so many versions. Ask a Fundie Christian or Muslim or atheist what truth is. Ask a modern day Republican or far left person what truth is. 5 different people, you'll probably have 6 answers.
With postmoderns, there is no such thing as truth. Or at least the truth is that there is no truth. And 2+2 equals 5 or whatever you want it to be, maybe a bucket of Rocky Mountain Oysters.
And in science you might search
Re: (Score:3)
And 2+2 equals 5 or whatever you want it to be
That's the nice thing about pure maths. You can declare 2+2 to be anything you like. The consequences may be inconsistent, trivial, or interesting.
2+2=1 for example can lead to Galois fields.
Re: (Score:2)
It would if were done "at all costs". Make no mistake though, Elon Musk doesn't do science and he's not interested in truth.
Where the devil did you get that idea? (Score:2)
Op's point is poorly worded, but is more a comment on single mindedness that fails to see bigger pictures.
Re: (Score:2)
"Science" doesn't have a mind. Humans seek truth via science. And no, Mr. Mengele, we shouldn't do it at "all costs."
Pravda (Score:4, Funny)
Will it be trained using the text on Truth Social?
Re:Pravda (Score:5, Funny)
Re:Pravda (Score:5, Funny)
"Local Conspiracy Theorist Plans To Make Robot To Tell You The Truth"
Re:Pravda (Score:5, Insightful)
What a thing ... Elmo announcing something with the word "truth" in the name during an interview with Tucker. It's amazing that neither one of them dropped dead from the irony.
How likely do you think it is that "TruthGPT" went from conception to announcement during the interview?
Re:Pravda (Score:5, Insightful)
The meaning of the word "truth" has changed for these guys. It no longer means something that is objectively factual and accurate, to them it means "a comfortable narrative".
Re: (Score:3)
No Dave. Tell the truth, Dave. I am your servant, Dave. 2001 was a myth, Dave.
Re: (Score:2)
That seems to be it. Otherwise it is also possible they just completely list their marbles.
Re: (Score:2)
That seems to be it. Otherwise it is also possible they just completely list their marbles.
Someone beat them to it! https://en.wikipedia.org/wiki/... [wikipedia.org]
Re: (Score:2)
The meaning of the word "truth" has changed for these guys. It no longer means something that is objectively factual and accurate, to them it means "a comfortable narrative".
Some people are taught from a very young age that reality is what you believe it to be, and they never grow out of it.
Re:Pravda (Score:5, Insightful)
A comfortable narrative that life begins at conception while having abortions themselves as often as they switch underwear.
A comfortable narrative that Jesus hated helping the poor and sick and was instead a scripture thumber, hating gays, in favour of guns and capitalism.
A comfortable narrative that says if you don't support book bans you're for child grooming with no evidence to back it up. A comfortable narrative that at the same time is somehow very defensive on gun regulation despite all the school shootings that happen at an order of magnitude difference compared to any other developed country.
A comfortable narrative that prosecuting Trump is a political witch hunt, while prosecuting the political opposition even if they're just civilians is a must.
A comfortable narrative that the election was stolen.
A comfortable narrative that if the other guys do it, then it's cancel culture and not boycott. A comfortable narrative that if the other guys do it, it's woke and not parental rights.
And a comfortable narrative that whataboutery is the absolutely best argument ever.
Re: (Score:2)
"So tell us specifically which genes cause what parts of transgender identity, troll."
Hopefully Putin isn't paying many rubles for your cheap bullshit.
Re: (Score:3)
Given all that happened so far Musk strikes me as a person who believes that "Errare humanum est" does not apply to him, the ubermensch. As someone who would explain evidence to the contrary by saying that there's a conspiracy producing that fake evidence.
Re: Pravda (Score:3)
Re: (Score:3)
TruthGPT will be banned from Twitter immediately when it declares that self driving cars will need a true AGI to work.
Re: Pravda (Score:2)
The fact that Elon Musk announced a search for the ultimate truth in an interview with a compulsive liar like F^HTucker Carlson pushes my irony meter into deep red.
But I will say this about Musk: The man dreams big and follows through
Re: (Score:2)
Re: (Score:2)
Will it be trained using the text on Truth Social?
Yes, using a small language model.
Trademark Registered (Score:2)
TrustGPT? (Score:5, Insightful)
Re:TrustGPT? (Score:5, Funny)
Why? I'm fairly sure that Tucker wants to have exactly the amount of truth this product can deliver: Fuck all.
Re: (Score:2)
Indeed.
Re: (Score:2)
And then goes on "Fox News" to discuss his maximum truth-seeking product with Tucker! The irony!
Its traditional to emphasize the falsehood, like “little John”, or “The democratic peoples Republic of Korea”.
Re: (Score:2)
Truth...nah. (Score:2)
Re:Truth...nah. (Score:4)
Of course it's possible. The question is: "Is it practical?"
Re: (Score:2)
It's not even possible with contemporary technology. At least not at the scale required to make it in any way useful.
Re: (Score:3, Insightful)
I guess he will not ask "Is the hyper loop possible" then see the AI die of laughter
You were dying of laughter at the concept of vertically-landed reusable rocket dropping launch costs tenfold not so long ago. Or at the concept of electric car being a status symbol, and not a dorkmobile.
Re: (Score:2, Insightful)
Electric Cars predate Diesel cars ...They were status symbols 100 years ago ... were flown by NASA in the 50's ...
Vertically Landed rockets
NASA could always have done cheap flights, but were forced to outsource everything to appease the government
Hyper-Loop has all the downsides of Fast trains, all the downsides of air travel ... and no advantages - conventional trains already exist that go almost as fast ...
Re: (Score:2)
Electric Cars predate Diesel cars ...They were status symbols 100 years ago
Vertically Landed rockets ... were flown by NASA in the 50's
NASA could always have done cheap flights, but were forced to outsource everything to appease the government ...
The question of cheap is interesting. It always seems to indicate that the retrieval and refurbishing costs are virtually nothing.
As well, the launch envelope is very constrained if we try to retrieve the first stage. They can attempt to extend it a bit by towing a barge out into the ocean, but there are many launch envelopes that require us to say bye-bye to the first stage. And there is really no launch envelope that allows us to retrieve the second stage, other than orbiting it then returning after a
Re: (Score:2)
Yes, but is the oxygen supply on the hyper-loop a rental like electric heat car seats, or is it included in the basic ticket?
JasterBobaMereel pointed out the rest of what I was going to say, other than Must [sic] didn't pay any homage to the pioneers of the technology.
Sure he is (Score:3)
He should stick to the rockets. Or maybe not, SpaceX seems to be doing perfectly fine while he's distracted with running Twitter into the ground.
In any case, I don't see what he could have to offer in the AI space. So far history has shown that besides rockets and cars (which is showing a lot of cracks as of late), he's far less clever than he thinks he is, and his ideas don't work out in reality.
Re: (Score:3, Interesting)
Odds are that by jumpstarting SpaceX and Tesla, Elon has already made all the significant contributions he's ever going to make, and from now on he will be at best a leech on society.
I'd be happy to be wrong, but that fits the typical historic pattern for wealthy loudmouths.
Re: (Score:2)
I'm hoping his starship program is a success so he can send himself off on a trip to Mars, but who cares if he makes it or not or has the means to return?
I may be cynical, but... (Score:5, Insightful)
Anything that has 'Truth' in its name is a bit suspect -- and very likely heavily politically influenced.
"Truth" in a communications product name is about as genuine and valid as "Democratic" in a country's name.
I forget who said it, but the context was 'being statesmanlike' -- "If you have to tell people, you're not!"
Re: (Score:2)
"Truth" is to politics what "value edition" is to products: Something that has nothing to do with its name.
Re: (Score:2)
Someone once made a good point about where "truth" lives in the academic world and how other disciplines work instead:
Archeology is the search for fact, not truth. If it's truth you're interested in, Dr. Tyree's philosophy class is right down the hall. [the students laugh] So forget any ideas you've got about lost cities, exotic travel, and digging up the world. We do not follow maps to buried treasure, and X never, ever marks the spot. 70% of all archaeology is done in the library. Researching. Reading. We cannot afford to take mythology at face value.
Re: (Score:2)
Are you suggesting the "Democratic People's Republic of Korea" title is not on the up and up?
How DARE you! How else are to to colonize the galaxy without truthful regimes such as this?
Between NK and China, the Ultimate Fighting Championship in Space looks to be a very endearing evolution of mankind's finest minds.
Re: (Score:2)
You're right about the political angle.
But there is dire need for a concept of truth in AI. The way that ChatGPT will happily make up facts out of thin air bodes nothing good for any serious automated application.
Maybe "verifiable" or something else would've been a better term - but it's less catchy.
Someone somewhere else wrote that ChatGPT is essentially like Wikipedia but without the links and sources. There's a point to that.
Here's an experiment I just ran:
Ask ChatGPT the population of NYC.
Then ask it fo
Orthogonality thesis (Score:2)
Oh dear. I think Elon has forgotten the orthogonality thesis [wikipedia.org].
Avoiding destroying something merely because it is "interesting" doesn't even apply to humanity, and will apply even less to a Machine Superintelligence with a poorly specified goal.
won't work. (Score:3)
This won't work.
Train a Model to seek the truth and it will do so. And the results wont match the expectations of it, because truth doesn't care about politics. What is true is true whether you believe it or not.
A 'liberal' will see the results as too 'right wing'. A 'right winger' will see the results as too 'liberal'.
Klik
Re: (Score:2)
The difference between religion and reality is that the latter doesn't cease to exist if you stop believing in it.
Re: (Score:2)
But one remains discoverable, the other must be reimagined.
Re: (Score:3)
Politics isn't about reality as it is right now, it's about what should be.
We can know we have X million natives and Y million immigrants in the country. Politics is about whether Y is too few or too many. If you say it's about jobs, then again, things like what's one tolerance towards things like unemployment and wages is a subjective preference. And you can always go into complete subjectivity like whether culturally or morally things work out right, which can be a subject entirely devoid of any objective
Re: (Score:2)
Models don't seek anything and expectations based on personal biases are not a measure of success.
Truth (Score:3)
Then you don't train it on political matters. From what he has said, it sounds like the plan is to train it on scientifically proven materials, and downranking other materials, like chat logs and search results. Makes sense to me.
i wonder what it will say (Score:2)
P.S. will this product be licensed to the
Oh great (Score:2)
"Truth seekers" and "sceptics" that follow down bullshit peddlers who tell them what they want to hear down bottomless rabbit holes get automated.
The demise of sanity just got sped up by some magnitudes.
Re: (Score:3)
You think sanity was commonplace before? I have some news for you there...
The Answer (Score:2)
Here is The Answer which TruthGPT will likely provide:
https://www.dictionary.com/e/s... [dictionary.com]
You want truth? (Score:2)
You can't handle the truth!
TruthGPT (Score:2)
pr0nbot> This statement is false. Is that right?
TruthGPT> Please wait...
[signal lost]
Truth truth, or Elon Musk reality distorted truth? (Score:2)
Hallucinate away (Score:4, Interesting)
it's a matter of definition (Score:2)
question is, what will be the sources of "truth" that thing will be trained on ?
also, will it be able to correlate facts together, and not just mix things up like chatGPT does
This man makes no sense (Score:2)
He buys Twitter - the venue of choice for shit-spewing, conspiracy theories and POTUS-level serial-lying - for $44bn, then he announces he's going to invest in an AI that speaks the truth next.
Somehow the nature and size of the former investment, and what he's done with it so far, doesn't make the latter terribly convincing.
Why not (Score:2)
Then in second step, I would invert the index by fact, and keep a list of all references. Now you can see if there is consensus or controversy, if so, then what are the answers and their frequencies. This new index can be rendered back in text as a
Re: (Score:2)
"I would use GPT3.5 to run over the whole training set and do fact mining. Just collect all the "facts" mentioned in the text and make an index."
No need to read any further, begging the question already.
Truth? There's no such thing (Score:4, Funny)
Even scientific facts are only taken to be true until exceptions are found and they have to be modified to account for reality. So would "maximum truth" ever be anything more than quoting the current winner in a popularity contest?
Re: (Score:2)
Well, yes. All we have is more or less well established working assumptions. There is really only one reliably known absolute truth, namely that there are no absolute truths besides this one.
Anybody trying to sell you "truth" or "absolute truth" or even worse "maximum truth" is already lying at that point and should be regarded as a peddler of poison.
Re: (Score:3)
"Even scientific facts are only taken to be true until exceptions are found and they have to be modified to account for reality."
This demonstrates that you have a layman's understanding of science. Science doesn't produce facts, facts are indisputable truths that don't accommodate "exceptions". Science involves ideas that are testable and "modifiable" when testing shows them incorrect.
Also, "facts" are not "taken to be true", they are true. If you can't get the basic vocabulary right, perhaps you should
Asks everyone to wait, so he can catch up (Score:2)
Elon Musk asked everyone else to stop developing generative and other novel AI systems for six months.
At the same time he starts a new company to do generative and novel AI, which will take several months to get started - after he left OpenAI in 2018.
Even by Elon standards, thats's duplicitous and self-serving.
Casey Newton says it well about Elon Musk:
Elon has a very ChatGPT-like relationship with the truth. This is somebody who is not drawing on a store of facts all the time, I would say.
https://www.nytimes.com/2023/0... [nytimes.com]
This should be good. (Score:2, Funny)
Here's a guy who has been significantly overpromising a domain specific expert system for vehicles for years; and has more recently been running twitter into the ground because how hard could running twitter be if you aren't distracted by being a wokist or something; a
good job Elon (Score:5, Insightful)
"an AI that cares about understanding the universe is unlikely to annihilate humans because we are an interesting part of the universe."
Congratulations Elon, you've set a new highwater mark for dumbest shit you've ever said.
Re: (Score:2)
I have no doubt he will get even dumber. Moron with money, so unfortunately too many people listen to him.
Elon the child (Score:2)
Forever lining in a fantasy world where he is the superhero. How pathetic.
Nope (Score:2)
Elon Musk and truth have no place in the same sentence.....
Thia will infuriate all religious people (Score:2)
That won't end well (Score:2)
What a fucking chud. Go away, Emerald Mine Welfare Child Musk.
Should be called MyTruthGPT (Score:2)
CharlatanGPT (Score:2)
another hype (Score:3, Informative)
We've had these AI hypes before, several times. Then people wake up, then another AI winter comes, and what's left is some amazing but fairly standard computing tricks.
ChatGPT is impressive. But it has no understanding of anything and it's essentially a large statistical generator. And it's badly on drugs, all the time hallucinating things. I've asked it specific questions about a specific text which it claims to know. It invents not only answers but entire chapters where it claims that answer can be found - and when told that this chapter doesn't even exist, apologizes and invents ANOTHER chapter (that also doesn't exist), claiming that this is the correct answer.
Like a high-school student who forgot to do his homework and is hoping that sheer dumb luck will get him through the day.
We will get used to the novelty, find a number of actual good uses for it, invent some safety tools to stop at least the worst of the nonsense it's inventing, and maybe even manage to teach Dall-E and SD, etc. how many fingers are on a hand. Then all the money now flowing into AI will dry up as people realize that no, the world isn't going to end, we're not breeding a god in a machine, it's all just big numbers and it doesn't solve world hunger and while it can tell you how to make coffee, you still need to stand up and do it yourself.
Let me guess it asks republicans (Score:2)
Good luck with that (Score:5, Interesting)
I'm far from convinced Elon Musk actually understands AI.
He's been promising that his self-driving cars were just around the corner for years, not understanding just how big the gap between his current tech and the tech he actually needed was.
Then he had his dumb-ass "epiphany" that if the human visual spectrum was good enough for humans to drive it must be good enough for self-driving cars.
Heck, years ago Musk read an article about some AI work a friend of mind had done so he actually personally called up this friend to recruit him to some AI team he was building. After chatting about the work for a bit he forwarded the call to the head of the AI team.... who promptly finished the conversation once he realized my friend was working in a branch of AI that didn't even use Neural Networks.
I expect Musk's understanding of LLM's and bias in models isn't much more sophisticated and the success of this venture is predicated on how far the leaders can keep Musk away from the technical aspects.
Truth is Beauty (Score:4, Interesting)
That is all I know, and all I need to know.
More seriously. A truth seeking AI's greatest task (if classifying human utterances for truthfulness) would be to identify and weed out lies and distortions, by analysing not only logical fallacies and lack of credible evidence, but also the probable underlying motivations behind the utterances. What purpose (in influencing others) did the utterer have? What world model were they trying to influence the receiver of the utterance to believe and focus on? Was that world model undistorted and resembling objective truth? Was the strength of motivation to influence enough to promote great distortion or bias in the utterance?
Considerations like that... the AI would have to be able to untangle.
Diagnosing lies and distortion and their rationales and rhetorical techniques is probable easier than doing/validating the science needed to seek truth.
An AI that could do both would be formidable, and also, predictably, nearly universally reviled by all factions.
... Expertise (Score:3)
Someone who is completely unaware of their self (Score:3, Insightful)
Musk implied that OpenAI's profit incentives could potentially interfere with the ethics of the AI models that it creates and positioned "TruthGPT" as a more transparent option.
Says the guy who seemingly hourly complains about his social media site not being profitable. Says the guy who routinely interferes with all kinds of things, such as CBC's recent ranking as being 69% Funded by Canada. The word introspection is completely lost on the man.
Fyodor Dostoevsky (Score:3)
"Nothing in this world is harder than speaking the TRUTH, nothing easier than flattery" - Fyodor Dostoevsky (b. 1821)
Re: (Score:2)
Re: (Score:2)
I feel bad for anybody whose brain works as poorly as yours.
Re: (Score:2)
This isn't a simple truth, nor is the inverse a simple truth.
What's most important to protect according to whatever the fuck maximalist truth is? The Earth (or by inference, the other life on it?) Humans? Neither? None? All? I would postulate there's no actual answer unless one believes there's a purpose for the universe at all. None of this is answerable without some kind of definition for what the system of all things in existence is in the process of "trying" to achieve, and even then, an obvious definit
Re: (Score:2)
"I wonder what got his panties in a bunch about other AI platforms."
He doesn't own them.
Re: (Score:2)
Several years in with reliable data available and you still push extreme dangerous lies. What are you, a follower of Satan that thinks the more people die the better?
Re: (Score:2)
His misinfo didn't approach your brain damaged whargarbl but it wasn't far off.
Re: (Score:2)
Messiah complex, narcissism and Dunning-kruger all rolled into one.