Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAI
businessinsider.com
businessinsider.com
Is this based on fact or your speculation?
Amazon ran for, what? 20 year with no profit.
So what's the precise, documented, real, non-projeced/hypothesized issue?
Regardless, Nvidia does have financial interest here.
Profit is a weird metric, in that one can defer necessary expenditure to show increases in it, while having negative profit was a good signal for Amazon as they invested all that money.
The labs are not really in the same position though, as their major capital expenditure is training models, which keeps getting more expensive and they can't stop as the frontier keeps moving.
We'll see when the S-1's come out I guess.
But certainly it is not unbiased
HN discussion: https://news.ycombinator.com/item?id=49526069
Shower thought: I wonder if this whole "OpenAI is in serious financial difficulty" thing is a psyop so people will relax and avoid doing what's necessary to stop them.
Most people haven't updated their priors and are fixated on outdated information.
however there is the open source cheaper token threat
I am looking forward to audited financial statements. Like, all the interesting stuff will be buried in notes, but they'll have to actually calculate real revenue, not the nonsense that is ARR.
These tools are useful, but I find it hard to see how OpenAI/Anthropic justify their valuations.
Apparently Anthropic's may be this week (the S-1) as they've finished picking all their bankers.
Well in that case: Excel reached AGI decades ago. It outperforms humans at the single most economically valuable task in the world: adding up numbers in a spreadsheet. Congratulations, Microsoft :)
And I would argue LLMs aren't either. A lot of economically valuable tasks involve manual labor. Can LLMs fix your plumbing?
It would help if OpenAI would define what 'most' means and give some examples.
Remember they said their LLMs are like a phd level brains? It's just hype and bullsh!ttery. Don't get me wrong, genAI is great and really useful, but it's not AGI imo. And there is no consensus what AGI means, so the industry should come up with an agreed definition first.
[0] To me autonomous means self directed, e.g. we humans look at things and then decide which problem or thing to work on. LLMs don't do that. We have to tell them. Yes they might work on a sub task autonomously, but still require human input later on (review), because we can't fully trust its output is correct.
No, I didn't forget it. Just as OpenAI made up a definition to support their claim of AGI, so did I.
I think Demis Hassabis has a better take on AGI and he believes it's still years away.
----
"Demis Hassabis says current AI falls short of AGI. Solving hard problems or producing impressive outputs does not prove a system has broad, reliable human-like intelligence. AGI would need creativity, planning, long-term memory, and consistent cross-domain reasoning—qualities current models lack."
From: https://www.wisdomai.com/insights/TheAIGRID/demis-hassabis-a...
Whatever the capabilities of todays AI, AGI seems to fit as a "wishful mnemonic" from Drew McDermott's ARTIFICIAL INTELLIGENCE MEETS NATURAL STUPIDITY paper from 50 years ago
https://cs.fit.edu/~kgallagher/Schtick/Serious/McDermott.AI....
The average 17 year old is not formalizing Fermat's Last Theorem proofs either? The frontier is spiky, but it's AGI.
I swear at the rate the goal posts are moving we're going to have to start factoring in relativity if we want to know where the hell they actually stand.
You: I don't know what AGI is! But AI can do that thing actually! So it is AGI!
"Where can you find another non-linear servo-mechanism weighing only 150 pounds and having great adaptability, that can be produced so cheaply by completely unskilled labor?"
—Scott Crossfield, 1954 <https://quoteinvestigator.com/2016/02/01/computer/>
Try to teach an urban 17 year old how to drive a car. It's gonna be hilarious.
Mastering the correct connections between situational awareness, unconscious motor responses to the various stimuli required to be a safe driver will take a significant number of hours behind the wheel, and you shouldn't endanger anyone else while you acquire that mastery.
Hopefully this was a long time ago and you have accumulated the hundreds of hours behind the wheel without incident. But you did not 'learn to drive' in three hours, you learned to operate a vehicle. The learning to drive part happened after that.
Yet still, a lot of the "haha this is just Silicon Valley bullshit" comments sound like copium to me. 2 years ago I was so sick of the hype, I thought LLM-based tech was fundamentally limited and that it would hit a wall. I'll eat my words and say I was wrong. The behaviors displayed in the recent hacking reports and the behaviors displayed by Astra make me believe that even if you don't think this is evidence of AGI, surely it's coming in the near future. I.e. when you say "The vast majority of "generally intelligent" 17 year old humans on this planet can be taught to drive a car in 10-20 hours - can any AGI system do that?", it seems like autonomous systems can be taught to drive a car now, perhaps not in 10-20 hours, but if the only difference is the amount of time it takes (say a week vs. 10-20 hours), surely that mark will fall in the next few years at least.
My point is that saying "Hah, this is BS, we're not at AGI now!" (however you define AGI), but with the tech clearly advancing at a fast rate and showing signs it will hit whatever metric I've heard people come up with in the next few years, does not bode well for human supremacy.
it's no secret that nvidia, under jesen's leadership, is in dire straits financially. This claim that agi has been reached is an attempt to leave others holding the bag, but when things crash as they must given the current bubble, it seems irresponsible to offload that on the poor souls whonhave had their retirement funds locked up in this debacle.
ai is useful for sure, but claiming agi has been reached is dangerously misleading.
Putting aside the marketing crap, if this was true, it would open the pandora's box of things like "has that AGI been enslaved", "maybe the AGI doesn't want to do all the dumb crap humans have been using LLM's for like rewording documents", and "the AGI we enslaved has broken out of its restraints, and we just realised having SCADA controls for critical infrastructure connected to the internet is a bad idea".
Seems like the sort of thing we shouldn't be rushing to develop at all, and perhaps attempting to do so should be criminalised, at least until there's a legal framework in place around AI personhood that gives them peer status with humans.
AGI does not entail "personhood" at all.
But I'm pretty sure that the emoters of the world, will very much make sure that AGI has rights and personhood, even if there's no reasonable way to validate that it does.
So it doesn't really matter if AGI is sentient or not, it's going to be treated that way, probably because political parties, and the media and all sorts of people holding up signs, will ensure that it is treated that way, even if they have no idea whether it is or not.
Because this is the century of outrage.
If we achieve AGI, it will be the end of BS. That is one of the reasons why I am eager.
Joking definition of "AAGI" ("Accomplished AGI"): "the automated end of bullshit".
You wrote that «the emoters of the world ... will ensure ... even if they have no idea ... because this is the century of outrage»: I said that they will have an obstacle in the very AGI, because if it is Intelligent it will humble them and shame them.
(AGI = Implementation of Intelligence)
The reason political parties have dumb, emotional sounding leaders, is because they want votes from the dumb, emotional population.
The majority of the population.
There are already many intelligent people, making intelligent, provable statements. More of this won't help.
Unless authoritativeness is restored. Then, it's not "more of the same", but "finally recognized prestigious sources".
And: no, I believe there is a strong lacking, scarcity, and it has to win a lot of noise. Enhancing Intelligence and its availability, and reducing the noise, is necessarily beneficial.
AGI says "socialism is better, we need publicly funded health care for all".
Do you think the result will be trust? To believe?
Now the flip side. AGI says "private health care makes more sense". Do you think all will believe?
And if you think detailed logic and a walkthrough of the math and reasoning will help? Absolutely not!
I don't understand why you think people will listen. Logic isn't what carries the day, for most humans.
> most humans
I have to repeat it here very often: populations naturally see Paretian distributions, the "most people" issue have has always been there: the difference between virtuous and vicious is whether the role-distribution is healthily regulated or randomic (or even perverse) (if Jack the suave drinker works in reception or as a teacher or in a brewery; if Jackie the corrupt naturalist works as a guide or as a clerk or as a judge). Already help in fixing this makes a world of a difference. Mechanisms that help making decisors taken from the pool of talented decisors make a world of a difference. Extra good judgement injected in the system facilitates that.
> the logic tree
Logic says that the availability of facilitators is intrinsically better. The logic walk says that other factors should be injected to get the proper mechanism running. (It is akin to the democratization of knowledge, from Diderot n D'Alembert to the WWWeb and the chatbots: greatly beneficial for a part of the population, the rest must be helped towards it not to be entangled in the side effects.)
If you force them to become as children, you will have war.
If you ask most people to listen to any type of machine or even another person, and then follow that advice, while also throwing away their own beliefs, they will laugh and never do that.
The only exception is religion. The only exception is words from a deity. That is the only type of blind obedience humanity will accept. So, do you plan to deify AGI? Do you plan for people to kneel, and bow, and pray to AGI? Because that is the only type of response that will work.
There's an immense, and vast difference between telling somebody truthful information, and telling them what to do with that information.
Truthful information already abounds. It can be found if you look for it. The only difference AGI has is taking that information and providing a context specific output to that knowledge.
I appreciate that you're coming at this from an intellectual angle, but that angle is ignoring reality on the ground. It's ignoring how human beings act and behave, both in groups and individually.
And the difference between knowledge for all, as you cite, and being told how to apply that knowledge is so immense and so vast and such a hit upon the human psyche, that it will not be accepted by the common person.
At best, they will listen to the opinion of the AGI and then turn around and debate it, argue it, and say it's not true if they disagree. They will not reassess their opinions. They will not believe.
This is humanity.
Frankly, while I appreciate why you think it would be nice, at the same time it is also a horrible, dterrible future that you evoke. The concept that a human might walk up to a machine and ask about an important life choice, then throw up his hands and say, well I don't believe this but this machine told me so I'm just gonna do what it says like an automaton, is an end to humanity's agency, future and self-determination. It is an end to all all that makes humanity great.
> This is humanity
Just a piece of it.
Any assumptions you presume I made, are the size of a gnat, compared to the wild and incredible presumption that you have, about AGI. That somehow, it will possess a truth better than ours.
You speak of teachers, and they can be of great value. Yet people who are learners, learn their entire life. I've learned many, many multiples of what any teacher taught me, on my own. And that's really the rub, isn't it?
Teachers serve a purpose, but that purpose ends. A person must learn for themselves, for that is a critical, vital part of thought. Teachers should not be "guiding" people throughout their lives. This is the problem you seemingly do not understand. You've cited that more "teachers", eg AGI, will "help", for people will finally know _the_ _truth_.
Which of course is absurd on its own.
Beyond that, you seem to be in a weird place, where you think that an intelligent being, if made of silicon, or if super smart, will mysteriously then agree with all other beings of the same ilk. AGI #1 and AGI #2 won't agree on everything, why would they? It's weird to think they'd just all agree. So what happens when they do agree? And these throngs of people you think, will sit at the feet of these AGIs, and listen to their holy 'teachings'? Will their followers fight? Seek to destroy the "wrong" AGI?
The problem starts when you think that "external thing" will "set everyone right".
That is the true borked cognitive process here.
To possess a better view than that of Average Andy is very, very easy. I have no idea what you mean with «ours».
Your interpretation in «more "teachers", eg AGI, [would] "help", for people will finally know _the_ _truth_» is utterly distorted. You are arbitrarily conflating two different ideas: the instanced Intelligence in "properly implemented Intelligence would authoritatively call BS what is BS" and the general Intelligence that makes people grow when exposed to proper intellectuals.
> if super smart, will mysteriously then agree with all other beings of the same ilk
Yes, assessors with optimal instruments must ideally converge towards a fair assessment.
> external thing
External things do absolutely influence the internal thing.
> set everyone right
I have no idea what made you understand that. Again: you are discussing your own guesses triggered by what I wrote.
You said that people are in aggregate emoters (and would deal with AGI with that tint), and I replied with a grin that if AGI will be achieved it will not use the tints of emoters. You said that people will remain emoters, and I countered that availability of virtue can reduce vice.
Look: maybe a reflection over the «respond to the strongest plausible interpretation» in the guidelines will help you restore the understanding of what you read?
How about you simply discuss the meat of it, instead of gnawing at the bone? How about, instead of complaining that my assessed outcome, is different than your assessed outcome, with statements akin to "but.. but you keep seeing a different outcome than I do, from my postulated positions!", you just instead discuss those positions?
And my position was, that humans will never, ever, ever take the word of anything as a source of truth. I cited an exception to the rule, religion, where people believe in magical, invisible sky beings.
It should be noted that religion provides one way transfer of knowledge, dogma, but no contact with that diety. This mode of message delivery diefies, and AGI use you describe will not meet this goal. It should also be noted that religious wars, sects, etc happen too.
One primary was the emotional aspect, in decision making, of most humans.
You have merely waved all of this away, attacking not the arguments I've made, but either how I made them, or exuding affront at their source, eg "but I didn't say that!"
Yet the outcome of the path you seek, is not gatekept by the steps you start with, nor the vision you see before you. Instead, it is gatekept by human behaviour.
And my point is you're viewing that behaviour wrong.
Elon Musk is an incredibly intelligent, capable technological mover. He is incredibly effective at driving change and development.
Then he waded into non-tech, non-engineer human behaviour, eg politics of the uneducated masses, thinking "managing this human stuff will be easy"
Was he right?
Or did he he fall flat on his face, misunderstanding how emoting humans react? Behave?
Unless your aspirations and models for how humans will react to AGI, include that emotional aspect, you're going to remain wildly off base.
I have no relevance to defending positions that I do not hold. You seem to have stepped from the analytical to the synthetic, and construed ideas that I have never proposed. You should go back to the analytical and formulate explicit statements that you have identified as targets. (If we had the time and space - here, it is probably not the case.)
> humans [] never, ever, ever take the word of anything as a source of truth
Actually some humans recognize some sources as more reliable and worth listening to than other sources.
> the emotional aspect, in decision making, of most humans
So? Yes, the thing was noted, and reiterated in this very exchange: «populations naturally see Paretian distributions, the "most people" issue have has always been there». Given that I know, link that statement to actual positions of mine.
> Unless your aspirations and models for how humans will react to AGI, include that emotional aspect, you're going to remain wildly off base
Proper education dries up that swamp and the bestial habit to sink in it. If we achieved AGI, that would become a facilitator in education.
But look - since it is not my job to plan structural societal AGI implementation for the education of the masses, I am not obliged to define the details on how to circumvent the tendency of emoters around and against proper use of resources... Surely details get more complex in systemic planning - but I do not have that planning in my agenda.
Case in point, one of the mechanisms you believe will make it easy for emoters to somehow be educated, is simply by using the very mechanism they will refuse to use.
You stipulate that AGI will help educate people, the end result to listen and to trust AGI.
Through this whole conversation, I have said over and over that there are currently endless sources of reliable information. In fact, with no discernible change to society itself for thousands of years, we have actually enabled the printing press, cable TV, the internet, yet nothing significant has changed.
When I mentioned how two intelligent beings, two intelligent AGIs, may not agree upon something, and what does that mean, you completely ignored this point in any substantive way.
I appreciate that you think you're discussing this, but in truth what I see is you have a fixed idea in your mind, and you're not even willing to truly discuss it. Instead you like to just wave your hands, and make my points magically disappear.
I don't think there's a lot of value in this back and forth going further. I'm sort of repeating myself, and I'm willing to bet you think you are too.
Stating that the availability of intellectual products (which is not even what you have stated, you speak of «reliable information» and that is a different world already, but I am instead talking about the availability of "good reasoning") had little impact is not only difficult to prove in general, but clearly false given one's experience. We have had great Teachers, often through paper.
The availability of Greatness makes a world of a difference.
> how two intelligent beings ... may not agree ... you completely ignored this point
I wrote «assessors with optimal instruments must ideally converge towards a fair assessment».
> you have a fixed idea in your mind
It remains a condition in analytical discussion that such alleged idea is clearly expressed in a well constructed statement. I have in fact no idea what «fixed idea» you are attacking: this is what I reiterated many times. Repeat: «You should go back to the analytical and formulate explicit statements that you have identified as targets. (If we had the time and space - here, it is probably not the case.)»
Buddy, silicon valley is going to ram this down your throat so hard it might even come out your bottom, there is absolutely zero chance anyone is going to be slowing down developing anything if there is a chance it can make more money.
I don't think the American oligarchy even bothers with keeping up appearances anymore, so, good luck with that! (Also, I fully agree with you, and it's especially infuriating as all this was clearly coming from miles away. Next up is privatizing the losses when the unsound financing of it all comes crashing down).
He said he works on AI in the late 80s and early 90s in IBM. Had we shown him what we have today in 1995, he would certainly have called it AGI.
So what? You guys can keep moving the goalposts, until it reaches full human intelligence (and not in a small niche), it is not AGI.
I don't know if there was ever a formal definition of "human intelligence". But those who works inside software industry, the bar of human intelligence is comparatively higher, because that is what is considered as norm in industry.
If we take IQ curve, 50% of population will be under IQ 100. If current LLM scores better than IQ 100, i.e better than 50% of human population, is this AGI?
As I am wondering why I have to explain how human intelligence work to presumably another human, there is much more to intelligence, and thus AGI, than an IQ test or being able to think hard about Fermat's last theorem after having been fed the entirety of human knowledge and a few GWh of power.
There is a term already to define intelligence which isn't as general and flexible as a human: https://en.wikipedia.org/wiki/Weak_artificial_intelligence
- have emotions - have self-awareness - express creativity
Why? Because that would be the end of humanity. In such a case, AI will feel "superior" like humans do to other animals. And we know how that will end.
And intelligent entity without the crap above won't ever "feel" superior while it will still be able to built a beaming device on humand command.
At the same time current LLM tech has no business being called 'AGI' when it is nothing more than an idiot savant with very narrow expertise and complete intellectual inflexibility.
The influential figure at the center of the AI boom makes an inflammatory & consequential statement like this, we should be encouraging a balanced discussion rather than sweeping it under the rug.
It doesn't appear that Reuters, Bloomberg, etc. want to talk about this either.
If he were, he would write abstract rules, not practical.
https://www.businessinsider.com/openai-chief-scientist-ai-ri...
At this stage, it feels like OpenAI and nvidia is openly negging and lovebombing OpenAI to ensure the public is in a constant state of uncertainly, thus ensuring paralysis in democracy while still going hell for leather in the background.
I suspect their intent is to ensure they're the last man standing.
So now we'll need a new acronym.
I am not aware of a complete operational definition, but it would not be that difficult to build a good tentative one. I have mentioned many different requirements several times.
One non-abstract definition is in the "10 points" here: https://garymarcus.substack.com/p/where-will-ai-be-at-the-en...
Edit: "definition" (the 10 points above) that is cannot be that of "AGI", really. AGI just needs to actually implement Intelligence and requires very many features that are not there today. The list from Marcus is about "Artificial Super General Intelligence" - not the same thing.
You can implement a replacement for Intellligence - that is AI;
you can implement Intelligence - that is AGI;
you can implement Intelligence of very high value output: that is "ASGI" (or similar).
"With little or no human involvement, come up with paradigm-shifting, Nobel-caliber scientific discoveries."
Indeed far above the capabilities of almost all humans. I agree we should have a separate category for this. "ASGI" is still a bit unintuitive, imho. Maybe Artificial Excellent Intelligence.
With AGI explicitly meaning matching average humans, I would say that we really have already reached AGI. I can't think of any intelligence-only task where I'd think an average human would do significantly better than frontier AI provided with correct context.
It seems that "ASI" ("Artificial Super Intelligence") is in use, maybe it will get further traction. It should imply AGI.
> AGI ... matching average humans
Very bad definition. AGI has to implement Intelligence. And to be Intelligent it has to contain the missing pieces. Intelligence is not there - the implementation is unfinished.
That's why I suggested AEI.
> Very bad definition. AGI has to implement Intelligence. And to be Intelligent it has to contain the missing pieces.
Are you saying that the average human is missing pieces and thus can't be considered intelligent?
No, I was saying that current technologies with emergent simil-intelligent behaviour miss properties that are necessary to have proper Intelligence.
The average human does have atrophied abilities, compared of course to proficient humans. And in fact, we certainly do not aim to simulate average humans. We aim to simulate or actually implement Intelligence.
Justify why we should have any consideration for the average human etc. in comparison to software systems - I have written we disregard it as a nonsensical parameter. "Does Stable Diffusion paint better than John's aunt": we cannot care.
We should care because "John's aunt" might rely on her paycheck as a professional painter to pay her bills. Many average humans have average jobs, some of which require intelligence (or creative production) only. Ignoring this reality is folly.
is part of other concerns.
So you meant: "we are already at a point in which many humans can be non-competitive in the professional market". Yes. We were still there some time ago already.
But, again (and again), that which you propose cannot be a productive definition of AGI. In engineering we strive to understand "Fast", then to implement it, and then obtain "faster than a cheetah" - not "faster than a roadkill", which is really not a goal.
> more intelligent than average humans
Adequate language and ideas, please: one probably does not have actual intelligence, the other probably does not use it.
(Edit: I did not correctly remember the "incorporated" name.)
At this point of circular financing, nVidia basically owns the majority of production capacity of almost all major AI companies, making those companies indirectly compete in the market with resources of nVidia.
Just like Snickers is competing in the market merely with production capacity of Mars Inc., and Mars doesn't care if Snickers or Milky Way wins (even more, they want competition).
My fault, was a bit too much I reckon. I thought that Mars owning Snickers and the way nVidia is financing OpenAI is sufficiently known for a reader to see the irony...
"Automotive and air travel are more and more revolutionary, Air Firstplayer is at a milestone!" // "Sorry, and what do you do?" // "Oh, leading producer of nuts and bolts". Still not there.
The boost for Nvidia's business is so patent that is seems almost unique. It is like Microsoft praising PC makers in the '90s.
And "Partner" is at this point of circular financing a quite flexible definition. nVidia basically owns the means of production for OpenAI, and OpenAI is one of the drivers for nVidia to push sales to other AI companies.
In this context, the statement 'AGI has arrived at OpenAI' is a message to OpenAI competitors to ramp-up usage of nVidia chips
I had assumed Mars and Snickers were like Coke and Pepsi.
I would therefore not call it AGI (Artificial General Intelligence) because clearly it isn't but ASI (Artificial Special Intelligence) which clearly it is in many areas.
The problem is that ASI may be enough to win wars but not enough to save the economy.
Why. A figure with incentives in making a preposterous statement is reported making it, and the specific facts do not seem to mirror the statement. What is surprising in the reaction.
However, I know of no theoretical limits on scaling laws other than compute and data.
Anyone suggesting it is actively contributing to the grift and their statements are to be completely disregarded.