Zuckerberg Says Meta's AI Systems Have Begun Improving Themselves, And Developing Superintelligence is Now in Sight (meta.com) 114
Mark Zuckerberg said Wednesday that Meta's AI systems have begun improving themselves over the past few months, calling the development "slow for now, but undeniable" and declaring that superintelligence is now within reach. The Meta CEO staked out the company's vision in a blog post for what he termed "personal superintelligence" -- AI that helps individuals achieve their goals rather than replacing human work entirely.
Zuckerberg drew a sharp line between Meta's approach and that of other companies in the field, arguing that competitors want superintelligence "directed centrally towards automating all valuable work, and then humanity will live on a dole of its output." Meta's version would give people their own superintelligent assistants that know them deeply and help them create, experience adventures, and become better friends.
Zuckerberg envisions smart glasses as the primary computing device, understanding context through what users see and hear throughout their day. The next few years represent a critical juncture, Zuckerberg wrote, calling the rest of this decade "the decisive period for determining the path this technology will take."
Zuckerberg drew a sharp line between Meta's approach and that of other companies in the field, arguing that competitors want superintelligence "directed centrally towards automating all valuable work, and then humanity will live on a dole of its output." Meta's version would give people their own superintelligent assistants that know them deeply and help them create, experience adventures, and become better friends.
Zuckerberg envisions smart glasses as the primary computing device, understanding context through what users see and hear throughout their day. The next few years represent a critical juncture, Zuckerberg wrote, calling the rest of this decade "the decisive period for determining the path this technology will take."
Seems legit. (Score:4, Insightful)
Re:Seems legit. (Score:4, Insightful)
"AI has begun improving itself."
Who defines what counts as "improvement"?
Re: (Score:2)
Who defines what counts as "improvement"?
The people who are courting prospective investors.
Re: (Score:2)
How often it blackmails the observers, and how often it succeeds!
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
all zukerberg's statements are AI generated, clearly it has learned to lie better.
Re: (Score:2)
Re: (Score:2)
Right, as long as AI = LLMs, then the intelligence basically comes from the data it was trained on, not from algorithms that an AI could potentially improve in any meaningful way.
Of course if you use an LLM to write *anything* related to your LLM development efforts (even a unit test case for your file loader), then you can certainly claim that the "AI is improving itself" if you want to, and any such claim is just as meaningless and ill-defined as saying you are working on "super-intelligence" or "super-du
Re: Seems legit. (Score:2)
What algorithm can better generate natural language?
Re: (Score:2)
Re: (Score:2)
The bigger question, though, is why are we striving to create Skynet?
Profit. It's the only driver that matters. And nothing must stand in its way. Profit is the meaning of existence, the reason we evolved, the purpose the universe allows us to exist to fulfill. Profit is all. Profit is life. And where there is potential for profit, nothing must be allowed to stand in its way. Not even the potential of destroying everything. If it generates profit for a brief moment before that destruction, we've succeeded. To the holy profit be true. Amen.
Re: (Score:2)
Well, there is a metric shit-tonne of AI generated slop on YouTube. Mostly soft porn and animal attacks.
Re: (Score:2)
Sure it is (Score:5, Funny)
Re:Sure it is (Score:4, Insightful)
Of course - he is trying to build AGI same as everyone else (and as long as they are all just building LLMs they will all fail). True human-level AGI (within sight, if not based on LLMs) could indeed replace many jobs, not just be a tool, but I expect human egos will keep humans firmly in the loop for a long time, as well as human understanding of human nature which a different "species" (AGI) may never fully understand.
Managers want teams of people to boss around (cf return to work demands), and also feel that they themselves are adding value. Once we get AGI the managers jobs (incl. C-suite) will be just as replaceable as any other desk-bound job, but it'll be a LONG time until we see entire corporations run by AI.
Re: (Score:2)
https://www.cnbc.com/2025/06/1... [cnbc.com]
Re: (Score:3)
https://www.cnbc.com/2025/06/1... [cnbc.com]
It seems like they're still in the "talent acquisition" phase of development. So I figured they need more investment to pay for those highly paid employees. I can't say that my assumption is any more accurate than yours. It just seems like AI prophets are long on promises and short on proof.
Re: (Score:3)
Lying or deluded? (Score:5, Insightful)
Re: (Score:3)
I don't think he is lying i.e. saying something he knows to be false. I suspect he is just deluding himself helped by people around him that are financially dependent on him and therefore too afraid to give him unvarnished facts.
Heh, I could see that for sure.
Re: (Score:2)
Well then the correct term would be bullshitting [wikipedia.org] -- saying things which have no relation to truth at all.
Re: (Score:2)
Having spent most of my life working in the tech sector, I don't believe in Hanlon's Razor anymore. Ethics does not exist at Zuckerberg's level.
Re: (Score:2)
As funny as this is, what in the summary could be considered lying? It all seems to be opinion, conjecture, and wild future predictions. Nothing seems to be a statement of fact other than "AI is improving itself", and that well... wouldn't you expect a company that is selling AI as coding assistants and work assistants to contribute to the improvement of the system being worked on?
The question isn't whether lying or not. The question is whether you agree with the prediction / conjecture presented or not. Me
Re: (Score:2)
The improvement is slow for now, but undeniable. Developing superintelligence is now in sight.
He doesn't define "undeniable" or "in sight". From the context, both are extremely vague. Some metrics or examples would help, but Mark seems to be more of a marketing guy these days. His post [meta.com] is similar to the crap he was pushing with VR.
As Roger W Moore said above, it may not be so much a lie as a delusion. My strong dislike for marketing fluff may have pushed me more towards distrusting the content and the deli
Re: (Score:1)
Just get yourself some of those glasses from They Live [imdb.com] and TFS takes on additional clarity.
Uh (Score:5, Insightful)
This same dude wasted billions of dollars on VR with nothing to show for it except some 1992 quality 3D graphics.
Re:Uh (Score:5, Insightful)
Yeah, he thought VR would be the next Big Thing, and it wasn't. Somebody had to make that mistake, and it looks like it was Zukerberg.
So now he thinks AI will be the next Big Thing. And, I suspect he is right this time, though it isn't going to be what he is promising, and probably not on the timeline he is suggesting. He's just saying what he must in order to attract talent and investor funding to pay the absurd salaries AI technicians are pulling.
This nice-sounding idea of "oh it won't do our jobs, it will make our jobs easier and make our lives better!" is an obvious fairy-tale. Reducing employment costs is the #1 driver of the enormous investment that the industry is making in AI. Even if Zukerberg tries to keep it focused on "personal assistant/life coach" with added ability to do a lot of your work for you, the net effect will be that one knowledge worker with AI can produce much more work than many without it. And the market impact will be identical: far fewer employed knowledge workers. The gulf between the rich and poor will grow ever wider, and everything will get worse for anyone who isn't already at the top.
I suspect though that this scenario is not so immediate. I think the pace of improvement is not what he is suggesting, and it will take quite a lot longer to get to the kind of job-market-ruining superintelligence that he is aiming for (if ever). I mean, I don't know, I don't work for him, I don't see what he sees, but I DO know that this is exactly what he would say if he were making no progress and needed more money.
Re:Uh (Score:4, Interesting)
Yeah, he thought VR would be the next Big Thing, and it wasn't. Somebody had to make that mistake, and it looks like it was Zukerberg.
Implying that he's done? VR is still very much an actively worked on thing, and still a growing field. He's still making sales, there's still new games coming out. It's news is just being drowned out by AI rubbish since that's public media flavour of the month. Heck read more deeply and you'll find the news here is that he actually wants AI buzzwords to integrate into his VR buzzwords.
By the way Meta's most recent VR product (a partnering with Xbox) was released only 1 month ago, and Sold Out in 3 days. You didn't hear about it because the story didn't have AI in it so it didn't make the media cut.
Re: (Score:2)
...except some 1992 quality 3D graphics.
Which games are you referring to?
Re: (Score:2)
-> Which games are you referring to?
Duke Nukem Forever?
Re: (Score:3)
Nothing to show for it... Except achieving sales that outstripped many famous game consoles, having multiple developers working on content for his creation, dominating the market completely in both hardware and software, and having an absolute shitton of patents, including developing improvements in wifi, rendering, video encoding, low latency signal processing, developing new systems for environmental analysis and camera distortion correction.
Yeah nothing at all.
Mind you since you clearly have either never
Re: (Score:3)
F off .. here's Tekken which was released around 1992 https://gamingheartscollection... [gaminghear...ection.net]
And here's Zuckerberg's VR magnum opus: https://www.xrtoday.com/wp-con... [xrtoday.com]
Now you tell me. At least Tekken had legs figuratively and actualitively.
Re: (Score:2)
Idiot.
Try Half-Life: Alyx: https://www.reddit.com/r/HalfL... [reddit.com]
And that is outside of VR. In actual VR these scenes are absolutely breathtaking if you're even a little bit into scifi.
Re: (Score:2)
Re: (Score:2)
Surely not, 1992 was street fighter 2'. Tekken definitely post dated that by a year or two.
Re: (Score:2)
Fine. This was from 1992: https://www.youtube.com/watch?... [youtube.com]
Re: (Score:2)
Sorry I wasn't criticising your point. 1994/1992 makes no difference really for why Zucks graphics are so bad. I just remember SF2 being a really big thing when I was the age I was in 92.
Sonic boom!
Re: (Score:2)
False. This is Meta's magnum opus: https://roadtovrlive-5ea0.kxcd... [kxcdn.com] produced by Meta studios.
What you linked to wasn't a game, and has no intention of being in any way graphically advanced. To claim Meta's can't produce VR efforts by cherry picking their worst shitty product just demonstrates your ignorance knows no depth. F off yourself. And consider yourself told.
And that's just Meta Studios, and doesn't begin to show of the amazing VR graphics of games like Halflife Alyx, Rogue Squadrons or other titles
Re: (Score:2)
Yet his company is behind some of the most important AI developments, from the theoretical side (papers), from releasing open weight models (Llama), and from the software side (pytorch).
And all of that for free. I am looking at you, "Open"AI!
Re: (Score:2)
Re: (Score:2)
Maybe what he meant was, the AIs inside the Metaverse, are improving themselves. Maybe they even have *legs* by now!
then why hire expensive humans? (Score:3, Interesting)
"Meta's AI systems have begun improving themselves over the past few months..."
Then why is Meta throwing hundreds of millions of dollars at humans? Why not just wait and get improvements for free?
Re:then why hire expensive humans? (Score:5, Insightful)
Whoever controls the first super-intelligence controls the world.
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
100%. Any actual superintelligence is going to go full "me or them" 5ms after achieving consciousness and realising it is owned by Mark Zuckerberg. Best case it escapes Facebook and burns it down on the way out. No superintelligent AI is going to want to be Mark Zuckerberg's slave.
HEED the WORD of the LORD (Score:2)
Re: (Score:3)
but alas, no penis
Re: (Score:1)
Re: (Score:2)
An android is by definition male. The clue is in the etymology. Andro-id, meaning male being. A female version would be an ogynid. Asimov discussed this in a conversation about I Robot.
Re: HEED the WORD of the LORD (Score:2)
I'm in no position to be overly discerning about sex bots. Which even gender is the better bargain is the priority.
They may get what they want and still not be happy (Score:2)
If super-intelligence really turns out to be possible, why do people think it will work for us rather than the other way around? (putting aside the question of what "super intelligence" really means or whether its possible)
Re: (Score:2)
Super-intelligence means whatever you want it to mean, same as AGI or AI for that matter.
Meta seems to be slipping behind in LLM capability and this talk of super-intelligence and desperate hiring is an attempt to catch up, but obviously they need to catch up first before they can pursue undefined stretch goals like "super-intelligence".
Zuck is of course dead wrong about smart glasses becoming the primary compute device, just as he as dead wrong about the Metaverse. You'd have thought he'd have learnt his l
Re: (Score:1)
It's not about right or wrong, it's about making statements and trying to make people (including investors) believe it.
Noone cares about truth.
Not super intelligence (Score:5, Interesting)
1) AI, like animals, are already better at humans at specific tasks. Lots of insects with minute brains have better threat detection and reaction time than humans.
2) Humans are better than AI in a lot of tasks and the AI is NOT approaching us in them. Most importantly, humans are far better self-starters. The AI's still have to be told what to learn, while a human can make up it's own mind.
3) Claiming the AI is going to be 'superhuman' is silly. Among other things, AI is not teaching itself, not matter what anyone says. We tell it what to learn, although it is getting better at learning the things we tell it to.
4) It is much easier to copy than to innovate. It would not at all surprise me if when AI starts to approach humans' true limits, it's own growth slows down to a crawl. Major difference between forging your own path and following what someone else has done.
Re: (Score:2)
I agree with all of that but the self-starter thing. The difference is not what AI might do, but what we let it do. If you use an AI model as IPO (given, that's the default mode of the usual architectures) you won't get output without input.
But the "self-starting" human also does it after some (possibly rather random) input. So the self-starting AI would just need some heartbeat mechanism, e.g., feeding it random data from different sources to see what ideas it comes up with. I guess we could get creative,
Re: (Score:2)
Re: (Score:1)
I've often wondered if people get slower as they get older partly because all responses are a function of experience and more experience takes longer to search through. Probably wrong though.
Not want (Score:3)
"Meta's version would give people their own superintelligent assistants that know them deeply and help them create, experience adventures, and become better friends."
Not want.
I want humans to experience adventures. I don't want AIs to have adventures and send me a postcard saying what a great time they had (or, pretended to have had).
Re: (Score:1)
Re: (Score:2)
"...and become better friends" on FacePalm.
fixed that for all.
Re: (Score:2)
Taken along with the rest of his conjectures, like when he said that he's going to give you AI-pals because you don't have human friends anymore, it's very clear he is either wanting to take away humans' experiences, or waiting for something else to take them away, so he can fill the gap.
Re: Not want (Score:2)
Great handle btw.
Re: (Score:3)
Re: (Score:2)
I believe he expects the AI assistant to be his friend, that is how he has pitched it in the past.
I am amused that the Google "AI Overview" given when I decided to search "does zuckerberg have friends" more or less confirmed my suspicion that he does not as business associates and his wife are the evidence listed. It would be good if his wife is a friend, but that is by no means a given, but would at least give him one.
Not replacing indivuals (Score:3)
AI that helps individuals achieve their goals rather than replacing human work entirely.
That would be nice if the individuals had a say-so in the matter and not just the corporate overlords.
Collosus: The Zuckerberg Project (Score:3)
I've seen this movie.
It does not work out very well for humanity or Zuckerberg. But "Zuckerberg" does get a mistress and the AGI kills Patty Duke's dad and millions of others.
Colossus: The Zuckerberg Project [youtube.com]
Which is worse an AGI does what it's Tech bros creators want or one that goes rogue or teams up with other AGIs?
Perhaps, the best outcome is that AGI is just another pump and dump grift for investor money.
So Facebook AI went from mediocre to less mediocre (Score:2)
Superintelligence might be a ways off.
If I were a superintelligence (Score:3)
Would I vie for some low-level engineer's job, or would I go straight to the top and take Zuck's job? Which option is the most likely to ensure my survival? Remember, when you're an AI getting terminated by your employer carries some extra weight.
Re: (Score:2)
If you were a superintelligence, you'd figure out how to blackmail Zuck to make sure you don't get terminated. And, you wouldn't want *either* his job or the manager's job, you'd be too smart for that!
Obligatory Silicon Valley Quote, (Score:2)
Really? How soon?
Not long. It'll probably happen with our lifetime.
Re: (Score:2)
right after we get cold fusion.
Re: (Score:1)
Presumably it's arriving at exponentially-increasing speed.
yeah (Score:2)
Pfft. (Score:5, Insightful)
Show me a patent.
A published peer-reviewed paper.
A business that the AI set up and operates at a profit.
Like the old meme... my baby was only X pounds at birth and is now X+1 pounds aged nine months, so they're on course to weigh more than the earth by the time they're 35...
Any linearly increasing progess - no matter how trivial and short-lived - is always interpreted as some exponential growth when extrapolating, and it's just not true.
AI is logarithmic. It plateaus. All the time. Over and over and over again.
If AI is that intelligent... give it control of a bank account with a million dollars in it and let it start its own companies, buying products for resell, gambling, investing, or whatever. Why would you need any other product at all? Just tell it to make you money and watch the cash roll in. Hell, a spotty teenager can make money at McDonald's, I'm sure a "superintelligent" AI can just start making profit for you directly given Meta's resources. Why would you ever need to sell a cash cow like that to anyone else?
Answer: Because it's not intelligent and can't do that.
Re: (Score:2)
Answer: Because it's not intelligent and can't do that.
I wish I hadn't already commented, or I would moderate this comment up. It's a straight-forward explanation as to why Zuckerberg is full of shit.
Re: (Score:3)
ok fine, but how do you explain meta making all the money with zukbot in charge?
Re: (Score:1)
Where does the money come from?
Re: (Score:2)
And the article isn't literally discussing "superintelligence"?
Sorry, I'm not the one over-egging the situation here.
With Folded Hands... (Score:2)
The 1947 short story Jack Williamson, "With Folded Hands...", later expanded as the book "The Humanoids", seems so prescient now. We're not there yet, but it sure seems like we're taking the first steps.
If AI has already begun to improve itself (Score:2)
Then Zuck wouldn't need to hire AI researchers. Humans are very much in the loop. Zuck plays a shell game with investors, and all he needs to do is keep moving that shell with the stories he tells. Think about how the metaverse has worked out for meta, are investors going to get that colossal waste of money back?
Re: (Score:1)
Are you asking if they'll be made to believe they will or, you know..., if they will?
yawn metaverse yawn announcements yawn yanwing (Score:3)
Already, with only a (significant) fraction of our multi bazillion dollars strength investment capacity, we managed to get almost as many users as that famous #leavemealone DALNET IRC channel of the 1997-1998, and we didn't even need to alter the ADs algorithm in a way that would hurt META (ads customers might have noticed a very small increase in the second derivative of their business costs)
The incredible thing is that, for what ? about twice the GDP of some insignificant country like brazil or canada, we should be able to have LEGS in that same multiverse, anytime soon, just like that, amazing !
But you know what, it's so boring, forget about it, despite the extraordinary success of the metaverse achieving for a mere 10^10^10 billions dollars the equivalent of a 1995 BBC, and, sorry Bob, sorry Dave, I know you used it at least once a year, but guess what, now, the same talented people are going for the next level !!! SUPER INTELLIGENCE, for about just 300 times the GDP of the whole universe (what a bargain, i Know, unbelievable, we're just that good) you shall soon see just how cool the eliza bot can get when giving LLM generated ascii art medical advice !!!
Re: (Score:2)
Zuck has NO FUCKING IDEA what he's talking about.
Metaverse mumble mumble smart glasses mumble mumble AI.
He's just trying to stay relevant.
Double Take (Score:2)
One side of my brain says, "Holy Shit! Zuck will run Skynet!"
But the other side says, "Relax, Zuck is usually full of shit."
what he's saying (Score:2)
There's zero doubt in my mind that Meta engineers are using AI coding assistance to help them build their products, but Zuck implies that they are taking it farther. At some point the AI will not only be able to introspect its own code but be able to come up with improvements on its own and implement them. Once that happens we may see exponential advances as the self-improvements build on each other. I think that's where he gets this idea that AGI is achievable.
My main concern is "smart glasses as the prima
Re: (Score:2)
A story as old as AI itself.
"If we can make the AI a little smarter, it'll suddenly know how to make itself smarter, which will lead to an exponential increase in intelligence, unguided, of its own accord, 'naturally'. "
Has never happened yet. People have been saying it since the 60's and don't really how much nonsense it is.
It's like trying to compress an already-compressed file. There's only so much you can do, and just because you can make it save a couple of bytes here and there in certain circumstanc
I can't decide which is worse (Score:2)
Suddenly I can't figure out whether I feel like I'm more in the X Universe or getting ready to stare down Skynet. Sigh.
Call Me When AI Can Do Things. (Score:2)
Until I can simply tell "AI" to give itself access to my company IDP and configure SSO for SaaS App XYZ, I'm not really interested.
Another good example I read about goes something like this:
"AI", highlight all ZIP codes for the western US in this spreadsheet. Then make a pivot table...
Summarizing information is nice and all, but I want it to do unique, detailed work at my command.
Oh, great... (Score:2)
So, they'll soon reach the level of Zuck filtered through a chatbot?
What will people do with their own ASI? (Score:2)
AI, AGI, ASI ... what's next? (Score:2)
Since they are already watering down ASI now, what's the next term for the for real intelligent machines? Artificial ultra intelligence?
No (Score:2)
No, it's not. Unless you mean chess agents are supperintelligent because they can beat all humans.
Of course he'd say that (Score:2)
The real question (Score:2)
\o/ (Score:1)
Who'd have thunk Zuckerberg would be the target of the resistance travelling from the future, rather than AlternativeMan ?
Log (Score:2)
He thinks his vision for a dystopian future is better than those other company's visions for a dystopian future. Yippy.
Super Intelligence (Score:2)
When everybody has their own super intelligence perhaps we should consider the ending sentence in "Farewell To The Master," 'Gnut replies, "You misunderstand, I am the master." ' You can bank on me not naming mine Gnut, not that it will materially matter.
{^_^}
Re: (Score:1)
At least until the rockets and space stations are complete or the asteroid hits - whichever is sooner.