OpenAI CEO Has No IPO Plan Due To 'Strange' Company Structure (reuters.com) 73
An anonymous reader quotes a report from Reuters: Microsoft-backed OpenAI, the company behind ChatGPT, has no plans to go public any time soon, Chief Executive Sam Altman said at a conference in Abu Dhabi. "When we develop super intelligence, we are likely to make some decisions that most investors would look at very strangely," Altman said. "I don't want to be sued by ... public market, Wall Street etc, so no, not that interested," he said in response to a question on whether he will take OpenAI public.
OpenAI has so far raised $10 billion from Microsoft (MSFT.O) at a valuation of almost $30 billion as it invests more on building computing capacity. "We have a very strange structure. We have this cap to profit thing," he said. OpenAI started off as a non-profit organization but later created a hybrid "capped-profit" company, that allowed it to raise external funds with a promise that the original non-profit operation still benefits.
Altman is on a whirlwind tour across the world, meeting heads of states of several countries, and was in the United Arab Emirates on Tuesday. He plans to travel next to Qatar, India and South Korea. While in Europe he got into controversy for saying OpenAI may leave the region if it becomes too hard to comply with planned laws on AI, inviting criticism from several lawmakers, including EU industry chief Thierry Breton. OpenAI later reversed the stance. "We did not threaten to leave the EU," Altman said on Tuesday. "We expect to be able to comply. There's still more clarity we are waiting for on the EU AI Act, but we are very excited to operate in Europe."
OpenAI has so far raised $10 billion from Microsoft (MSFT.O) at a valuation of almost $30 billion as it invests more on building computing capacity. "We have a very strange structure. We have this cap to profit thing," he said. OpenAI started off as a non-profit organization but later created a hybrid "capped-profit" company, that allowed it to raise external funds with a promise that the original non-profit operation still benefits.
Altman is on a whirlwind tour across the world, meeting heads of states of several countries, and was in the United Arab Emirates on Tuesday. He plans to travel next to Qatar, India and South Korea. While in Europe he got into controversy for saying OpenAI may leave the region if it becomes too hard to comply with planned laws on AI, inviting criticism from several lawmakers, including EU industry chief Thierry Breton. OpenAI later reversed the stance. "We did not threaten to leave the EU," Altman said on Tuesday. "We expect to be able to comply. There's still more clarity we are waiting for on the EU AI Act, but we are very excited to operate in Europe."
Retrofitting This into a Non-profit is Tricky (Score:3)
They're figuring out how Microsoft will get to do that.
Re: Retrofitting This into a Non-profit is Tricky (Score:2)
Or they just don't want to IPO for practical reasons. My coworkers and I don't want our company to IPO because we really don't like the way institutional investors effectively take over your company. You can claim that the Google founders control Google all you want, but watch what happens when institutional investors start shorting your stock because some random analyst with zero affiliation to your company predicted higher quarterly numbers than what ended up in your 10Q. And don't worry if it doesn't tan
Re: (Score:2)
Re: (Score:2)
But every startup must have an exit strategy. Either you get as big as Google, you get bought, or you go out of business/.
Re: (Score:2)
My employer only has one goal, and growing is only a means of reaching that goal. Going public would only make reaching it much harder, if not outright impossible.
Re: (Score:1)
super (Score:5, Insightful)
> When we develop super intelligence
While ChatGPT knows a lot of words and can mostly string them together, ChatGPT is far from intelligent let alone 'super' intelligent. What a bunch of BS. Altman purposefully conflates knowledge with intelligence, a string of words with intelligence. Can something that does not think even be intelligent?
Marketing (Score:3)
Re: super (Score:4)
At the end of the day, Altman is a businessman and businessmen do what business gotta do. Whether he actually believes in his own spiel is honestly irrelevant, since none of us can really tell.
It is surreal though seeing Altman, and other tech moguls basically adapt the televangelist marketing book. Preach a fire and brimstone apocalypse that can only be averted if you repent and pay for salvation through cold, hard cash. Whether it is the rapture or a malevolent AI, the spirit of the idea stays the same.
Re: (Score:3)
Indeed. Marketing lies by intentional misdirection. If they ever get a glimmer of AGI, I will be impressed, but it seems unlikely the ChatAI approach can do that at all.
Can something that does not think even be intelligent?
The only examples for general intelligence we have come with consciousness and free will attached. That is a rather strong hint that these three things are connected and may not be able to exist individually. At the same time, all efforts of artificial creation of either of these have completely failed so far and there really is no scientifi
Re: (Score:2)
A hint from where, the last fortune cookie you ate?
If you are of such superior intellect that you consider anyone a moron who considers a human brain to be a machine that runs AGI, why don't you explain it to us in simple terms?
Re: (Score:2)
Just requires working intelligence instead of mindless belief. Switch that on, suspend your belief in the machine God and you will see it as well.
Re: (Score:2)
Sounds like a typical example of a flimsy argument with no support to back it.
Re: (Score:2)
Consciousness is mostly just alertness and paying attention to stuff. Its also somewhat of a weasel word. We know what we mean, but good luck definings
Free will is an illusion. Its a phenomenological thing we experience, not something that actually exists in an actual brain, which most assuredly is still run by all the levers and dials of biochemistry as surely as a computer is run by the levers and dials of electronics.
GPT can already "Pay attention" to things (better t
Re: super (Score:3)
Free will is an illusion.
Obligatory: Why do we have prisons then? Why do we have penal law ar all? Why do we punish people for things which aren't within their power not to do?
Re: super (Score:2)
Illusion always seemed like a loaded word, or at least perhaps not appropriate for the discussion on free will. Illusion implies there is something real, which in this case implies that free will is real but differs from the human interpretation of the concept.
If nothing else, structuring our society around free will is likely better then the opposite. Jailing someone for crimes they may potentially commit is the stuff of classic dystopian fiction, but knowing parts of this community they would likely suppo
Re: (Score:2)
But we do give and take away travel visas to countries based on what we believe their people will do when they come over.
Re: (Score:3)
Re: super (Score:2)
Consider me a bit slow, just for the sake of argument. Can you spell it out for me like I'm 5?
Re: (Score:2)
He is pushing the common, but irrational "fear of the other" as a fake rational argument.
Re: (Score:1)
I guess the argument would be "We don't have the free will *not* to punish them"
It's a pretty flimsy argument.
Re: (Score:2)
Re: (Score:2)
"Betray" implies free will.
Re: (Score:2)
Re: super (Score:2)
Without free will, that's a non-sequitur. How would the counteracting work if people didn't have the ability to decide to change their behavior?
Re: (Score:2)
You have a nice set of opinions and conjecture there, but nothing scientifically sound. Physicalism is _religion_ not Science. Science makes none of the claims you make there. The fact of the matter is consciousness is not possible in the current Physical standard model. There is simply no mechanism for it. At the same time it can influence physical reality because we talk about it. The scientific state-of-the-art is that the whole thing is a big unsolved mystery.
And no, things are still moving glacially sl
Re: (Score:2)
Note: Of course the usual quasi-religious morons praying to the machine God will claim obviously human brains are just machines and obviously human brains create AGI. The scientific state-of-the-art does not support these claims.
What prompt was used to generate this text? I think this is rather clever.
Begins by preemptively deriding the very sort of mysticism it itself invokes. Followed by invoking unspecified science to push inherently unscientific concepts.
All we have are interface observations and that is not enough to conclude what must be going on inside without a GUT and all its parameters being exactly known. And we do not have that either, neither the GUT nor the parameters.
And the LLM hallucination at the end is genius. Even if a grand theory of everything were developed that predicts everything known it still wouldn't do you any good in terms of conclusions because you still can't know what you don't know.
NOT saying it's aliens of course but
Re: (Score:2)
I do not invoke mysticism. That is just you being disingenuous and manipulative. I claim the question is open. And it is. Unless you are one of these people that can only deal with "yes" and "no" and "unknown" has you in gibbering hysterics?
Re: (Score:1)
Re: (Score:2)
Define "intelligence" and "think".
Re: (Score:2)
What does it mean to have the power to imagine?
Re: (Score:2)
Imagination is similar to the process that a large language model follows when you prompt it to predict an answer.
Re: (Score:2)
Please explain your reasoning. I think you are saying that my brain executes a statistical model when I want to type this sentence. Can you provide some support for your conclusion?
What does ChatGPT do when you are not prompting it? Does imagine things in its spare time?
What happens in your brain when no one is prompting you? Do you think ChatGPT is the same as you in its off time? Does ChatGPT dream?
I don't claim to know exactly how my brain works. But I can say that ChatGPT does not work the same way my
Re: (Score:2)
I recommend reading "On Intelligence" by Jeff Hawkins; that book covers many of your questions quite well.
Yes, it is true that the brain executes several behaviors related to survival that are not applicable to large language models, yet. That's just a matter of implementation, though.
Re: (Score:2)
https://www.technologyreview.c... [technologyreview.com]
Why do you think AI is heading in the wrong direction at the moment?
That’s a complicated question. Hey, I’m not a critic of today’s AI. I think it’s great; it’s useful. I just don’t think it’s intelligent.
My main interest is brains. I fell in love with brains decades ago. I’ve had this attitude for a long time that before making AI, we first have to figure out what intelligence actually is, and the best way to do that is to stu
Re: (Score:2)
Yes, I'm familiar with his thoughts here, and also believe that in "On Intelligence", Hawkins deliberates about related topics such as the "Chinese Room" thought experiment, which asks "Where does intelligence reside?" While Hawkins acknowledges that latent intelligence resides in DNA as a series of information shaped by countless generations of evolution, he fails to acknowledge (or just does not do so explicitly) that the encodings within Large Language Models form the same kind of embedded intelligence
Re: (Score:2)
Define "intelligence" and "think".
The only thing the naysayers have left are word games. The won't dare cite testable objective tests because they are fully aware AIs have already blown through them.
Re: (Score:2)
Does AI imagine? What does it mean to have the power to imagine?
Re: (Score:2)
Re: (Score:2)
> When we develop super intelligence
While ChatGPT knows a lot of words and can mostly string them together, ChatGPT is far from intelligent let alone 'super' intelligent. What a bunch of BS. Altman purposefully conflates knowledge with intelligence, a string of words with intelligence. Can something that does not think even be intelligent?
I'm curious about the context, it's an awkward enough quote that I wonder if he was trying to make a joke. Something along the lines that the company would develop a super-intelligent AI that would take over the company and start making paperclips [lesswrong.com].
Re: (Score:2)
> When we develop super intelligence
While ChatGPT knows a lot of words and can mostly string them together, ChatGPT is far from intelligent let alone 'super' intelligent. What a bunch of BS. Altman purposefully conflates knowledge with intelligence, a string of words with intelligence. Can something that does not think even be intelligent?
He didn't give a timeline on that "super" delivery, and we went from driving the first Model-T to driving a lunar rover, within a single lifetime. And quite frankly, by the time we ignorant meatsacks develop super intelligence, the machines will have already determined who is expendable. And why.
I really don't know why everyone is so hung up on assuming it's going to take some kind of super-duper-hyper-awesome Intelligence to put the "I" in AI, but take a good hard look around you when you're at work next
Re: (Score:2)
Re: (Score:1)
Pathetic (Score:1)
10 billion will be wasted (Score:1)
Guaranteed the $10 billion will be fully spent within a year or two and there'll be nothing viable to show for it. It'll be like all the money Meta blew on the Metaverse or Uber on self driving. Or the government spending a $31 trillion loan yet we're not living in spaceships and there's still homeless people. Zuckerface spent billions and instead Apple, not Meta, had to show people what VR could be. Around the time of ancient Egypt, there was a wealthier kingdom called Nubia to the south. Does anyone know
Re: (Score:2)
Dude we already have a lot of warehouse robots and these advances will already make more. ChatGPT and co-pilot are both of high utility already. We're not getting AGI anytime soon but the recent demonstrated capabilities have shifted my opinion from maybe to definitely in my lifetime.
> Zuckerface spent billions and instead Apple, not Meta, had to show people what VR could be.
Quest was pretty good until they fucked with it. Now I can't say because mine is sitting gathering dust ever since they deactiva
Re: (Score:3)
Warehouse robots are mostly driven by expert systems and even the approximate control algorithms are more likely to be fuzzy logic than neural. No one is going to let some hallucinator loose with a forklift.
Re: (Score:1)
Re: (Score:2)
I'm guessing Nubia gave us n00bs. Also warehouse/assembly robots could be pretty useful. Especially if they're better than the ones we have now.
Re: (Score:1)
Zuckerface spent billions and instead Apple, not Meta, had to show people what VR could be.
All we have is a CGI commercial -- arguably "Ready Player One" has shown people what VR could be as much as Apple have.
'Strange' company structure (Score:1)
How charming.
Re: (Score:2)
How charming.
Yeah, he sounds insufferable.
Translated - "We're so much smarter than anyone, and all you peasants out there can't possibly understand the weight of our genius"
Easy fix... (Score:2)
Change name to "UnOpenAI"
To the Naysayers, this is not like crypto (Score:2)
Re: (Score:2)
Half the people here seem to be curmugeons still repeating the same tired "AI isnt real" arguments that might have made sense in the 90s when LISP machines and barely functional perceptrons where about as high tech as it got. I think a lot of these guys seriously think GPT is just some ELISA type bot with an ungodly number of prolog rules under the hood. Meanwhile neuroscientists are starting to discover actual human brains seem to arange knowledge surprisingly similar to how our modern convolutional type n
Re: (Score:2)
Re: (Score:2)
Re: To the Naysayers, this is not like crypto (Score:2)
ChatGPT has passed the bar exam, GRE, SAT and a score of AP exams.
Not-USA-ian here, so I nevwr had to, or wanted, to bother how such an exam looks like. But if ChatGPT passes them, this just goes to show how crappy and disconnected from the necessities of the real world that examination.
Not a week goes by in which not a blunder, or rather impressive fuck up, of ChatGPT is exposed, simply because it isn't smart; it's just good with words.
I agree with your premise, though, that this is, unlike cryptocoins, in principle useful technology and is here to stay.
Re: To the Naysayers, this is not like crypto (Score:2)
As it turns out, there is a lot more to being a lawyer (or at least a decent one) than just answering exam questions all day.
Or how being a software dev is more then just grinding out a higher Codeforces rating.
It is a tad insulting to think we can distill complex fields into a handful of parameters and prompts just to pinch a few more pennies off the payroll.
Re: (Score:2)
Not-USA-ian here, so I nevwr had to, or wanted, to bother how such an exam looks like. But if ChatGPT passes them, this just goes to show how crappy and disconnected from the necessities of the real world that examination.
Interesting domain experts would create unspecified tests to measure aspects of human intelligence while people who openly admit to total ignorance of the tests feel fit to confidently pass judgment about that which they have no information.
For all you know the GPT model was intentionally trained on data including similar answers to the tests knowing full well in advance their models would also be evaluated on this basis.
Not a week goes by in which not a blunder, or rather impressive fuck up, of ChatGPT is exposed, simply because it isn't smart; it's just good with words.
I know people expect computers to always spit out the right answers otherwise they assu
Re: (Score:2)
I don't think "AI" will blow over. However that doesn't mean that these LLMs that are being hyped up so much are the future either. Neural nets are great at getting you "sort of" solutions. Lots of 80-95%s. And the last 5-20% tends to remain elusive due to the black box nature of these models. You can throw more data at it, adjust the training parameters, try and augment the output in some way, but ultimately they keep falling short outside of weird random problems that they unexpectedly manage to solve (95
Re: (Score:2)
Re: To the Naysayers, this is not like crypto (Score:2)
Ehhh there are a lot of legal problems with generative AI tech like SD as is. Lora in particular just strikes me as just a straight up tracing tool that allows for users to submit and download trained models that are extremely close to the style of the affected artist. We talk about transformative uses but man some of it really edges closely into outright mimicry.
Does not really help that apparently in Japan there is a business that specializes in selling LoRA models built off of artists without necessarily
Heh. (Score:1)
Re: (Score:2)
Wants regulation, as long as they don't affect him (Score:2)
Non profits without a strong democratic structure beyond the ability of the board to change will turn into scams. Grifter pretending to be jesus.