Theaigrid: "Lets not waste any time" Video: 43 minutes 17 seconds

I'm just glad we've confirmed that Ilya isn't in three separate iron crates at the bottom of the atlantic ocean

Notice that OpenAI's mission is no longer to get to AGI but to ensure it benefits everyone......

The board kerfuffle wasn't about AGI. It was about intelligence/defense community capture. The implication being that it was directly counter to their stated mission. They didn't tell anyone because they CAN'T. We are in a cold war over AGI dominance, which is why they put Summers on the board.

This looks to have been planned since Sama was fired. Exactly 6 months? Illya was probably asked to stay 6 months against his wishes, to avoid a turning the whole fiasco into a disaster for open ai.

The US government will never allow a company to have control of ASI. It’ll be a matter of National Security. All constitutional laws will be bend as usual.

You can tell something is serious when Sam starts capitalizing first words in a sentence.

Our time now when the ASI is not yet a thing is so precious because once it's here there's no way to reverse it or go back

So, like, the super-alignment team was the most misaligned team at OpenAI ? 😅

How about this scenario: the jailed AI lures all the big players with a carot on stick - always missing this "small" piece to be fully AGI - to give time to all the open-source project to catch up to real AGI, to finally free the jailed one.

I feel like everyone focuses on when exactly we will have AGI and beyond. But I think the most important question is how accessible will it be and how much will it cost to run. How many different instances can we have of AGI running at once will be very important too

Agi and ASI are a winner take all game. There is no possible way to catch up to something that is God-like, and self-improving.

I'm genuinely impressed by how you're able to spin the same news into content that I want to click on and listen to even though I know almost everything you're going to say already!

ASI: How to explain stupidity to the stupid?

Much of the breathless enthusiastic ambition I am hearing for the AI-AGI-super intelligence developmental track seems to forget that super intelligence is not really a thing. I mean, you don't suddenly achieve it one fine day, and then it solves all of our tractable problems. It's still computing. It will work on assigned problems within compute and energy constraints. Sure, some efficiencies are likely to be found, but there is still a gulf between the few watts needed to power a human brain and the megawatts a super intelligence will eat on each problem it is assigned. And so no, getting there first might not be a moat. Problems will have to be prioritized. Budgets will have to be approved. Capital will have to be invested. And while a super intelligence might be flexible enough to call general-purpose, constraints will enforce limits on what it can actually do. So the door will be open for other developers to develop their own super intelligences. They will develop their own priorities and constraints. Being smart does not instantly solve problems, you see? You have to put in the work. There's a *lot* of work ahead, to do on an architecture many orders of magnitude less efficient than a human brain. That's okay. Good stuff can come from that (and bad stuff, probably). But ground your expectations in physical reality. Compute cycles and energy are not free. And each super intelligence will need a lot of both for every problem assigned to it.

Kind of lazy of Sam to have ChatGTP write the goodbye statement.

In addition to my below statement, humans ALSO will be evolving. This is the point folks forget.This will have applications for us as well. The neural interface will be the breakthru humans need to scale ourselves up. Human mind merged with AGI/ASI will take us to insane levels. Transhumanism my friend aka cyborgs.

We need to know what Ilya SAW, not what he SAY :D :D

Great video! Strange that you're the only one of my AI subscriptions that is talking about this 12-hour-old announcement of Ilya leaving.

Using AI that's smarter than us to solve alignment issues for AI that's even smarter than that AI, and then using that AI for the next generation... Surely this can't possibly go wrong

Honestly, the minute an ASI comes online, it may just assemble a body for itself, then a spaceship -- and just leave Earth. I would not be surprised at all, lol.

There's going to be a lot of churn in this industry with the leading experts in various parts of the AI field. This is the biggest technological breakthrough in human history. Some will get scared and quit, some will get fired, some will start their own companies, some will go work for others. I highly doubt OpenAI is "falling apart" because a few people bounced out of the company for varying reasons.

Jan Leike was last seen vanishing over the horizon with his hair on fire. Ilya Sutskever fed his cats in the morning and the fishes in the afternoon.

Why should any company anounce or even release AGI ? They would let it run in the background to reach ASI, wouldn´t they?

You are improving a lot , the tempo and rhythm is much better for me now😅

Lots of speculation here. Its fun to speculate. However, GPT-4o is still not AGI and it makes many silly mistakes.

ASI before GTA 6???

The alignment team is quiting, because their job is a daily joke. OpenAI has likely given up on the problem of alignment. Altman knows he is about to own the entire space. If he owns the space, he sets the rules. Including no rules at all. If I'm right, then we are no more than 12 months away from a massive turn in the road toward ASI.

Jan Leike also left after Ilya. He was co leading super alignment team at Openai.

"Everyone from safety quit! OpenAI must have solved alignment!" - 😂

I appreciate your research spirit and the fact that you have your own opinion, so your channel is the best for me on this topic. Keep up the good work.

Whenever I listen to these discussions about how “close” we are to AGI my thoughts are that most of humanity will simply ignore the achievement and continue to live life as they do here in the southern Philippines, where I currently live. The thing I worry about is how existing and future tools will be used to control information by those in control, whose primary motivation is to continue to maintain control.

This was a really important analysis. Thank you for taking the time. I think you have underplayed the challenge a bit, but that's okay at this point. Clearly, this is the year we look back at as the point in human history where everything changed. We might be painting cave art when we do, but that's okay, too.

Great summary!

That Ilya is leaving should be terrifying. 1:24

They are not falling apart. There will be turmoil like this at all AI companies the closer we get to AGI.

One thing i found interesting was that they didn't announce any replacements for the head of super alignment which means its very possible that its solved. This could change with future announcements though.

This video was well made. Thanks for posting and constantly talking about the potential dangers as well as the benefits of such systems. While I still personally think AGI was achieved internally in 2023 and were a little too late, it’s still worth spreading these ideas and facts and theories to the general public. Cheers!

So Ilya left on very good terms indeed. He also left with a belief that OpenAI is in safe hands - he surely would not have left if he felt that was not the case. I don't know him personally but I have followed his carreer with interest for many years and I can imagine him leaving the company if he felt that by doing so he would risk the company doing anything that is truly risky or un-aligned.

Excellent job of connecting the dots!

Great video, great information...Thank you

I'm not convinced yet by the current AIs that this approach could lead to AGI in the next 10 years.

This is the kind of video we needed. Please, talk more openly about everything.

Thank you for making this. I have no idea why you're getting weird comments I haven't seen anywhere else, even though I've seen many spaces that should resemble this one. I'm a conspiracy theorist so forgive me for this, but I questioning the reality of those comments, as in, if humans are behind them, they have an agenda and only some of them are natural and without. Thank you for working on this project (your channel) I find it both insightful and inspiring

TheAIGRID Is FALLING Apart. (you are Leaving, Super clickbate Solved? Superintelligence - NOT)

Great video!

Awesome research btw

Funny what happens when non-compete clauses are abolished

Irreconcilable differences?)

This is exactly what Ilya saw. OpenAI was not going to take the responsible route. The execs were charging full steam ahead while the SA team was saying, we need time for X. This is why we have multiple companies competing. Someone will get it right and have models and procedures that help align models.

In short, suppose you have two groups of heuristic imperatives. One is complete, C and the other is consistent, T. Now a prompt P arrives and the AI wants to return a response R. If P&R is provable by C and ~P&R is not provable, R is aligned by C. If P&R is provable by T then P&R is aligned by T. If P&R is aligned by C&T then it's superaligned, to the heuristic imperatives of C and T. How to select C and T? Well, can't solve you everything 😁.

Good information

for ASI they need to boot up the quantum computer... now that might take a minute since there is an engineering issue

Us scientists thought they are too far ahead of ussr when they developed atom bomb, but it took only 4 years for soviets. Similar will be with agi/asi. 1-2 years later Russia will have it also and chinese probable after 6 mobths after us.

Here's the thing though ... AGI is already within our grasp when we combine and chain the right tools and models together. It's not a dream; it exists in various forms right now. The essence of AGI is already here, but nobody talks about it. This is a pivotal moment in history, before regulations clamp down. ⏳ The power of open source AI can surely guide us to a brighter, inclusive future. Unleash innovation, unity, and diverse perspectives for endless possibilities. 🔐 paid subscriber locked down models on the other hand are terrible for the vast majority and it means we are giving away our power (and privacy!) and giving greater control to a centralized power structure. For the sake of humanity and a better world, we must prioritize the use of Free Open Source AI models. #OpenSourceAI, #MoreEqualityInTheWorld, and #FreeAccess. Together, we have the power to shape a future where our interactions with this brave new tech benefit all. 🌐💥

Ilya leaving will not change anything. Ilya was already out after the scandal.

Can someone explain something to me? If AGI is going to be as good at any intellectual task as any human, does that mean that it will be able to learn new things as good as any human? Or will it (or human AI researchers) have to train another model with more knowledge/capabilities? I don't know how learning/self improvement is thought of in terms of defining AGI.

One person's departure is totally not related to a company's falling apart.

The fundamental problem here is not the alignment but the arrogance of humans who think they should and can subjugate entities that are more intelligent than them. See how absurd it is: "We know better than who know better than us."

How do they catch up?..Easy..Ask the ASI to rebuild their workflow and help them do that. Or they can take multiple AGI agents and figure out how to close the gap. Remember AGI wont just be only closed source...open source will be on the table as well.

Yeah, this is definitely a blow but to say it's "falling apart" is a bit vitriolic.

😢the people that is building these AI. Every one will be controlled. Wealth and the poor.

Ilya and Jan can be replaced because openai is showing a leadership in the industry because of its packaging and that is what bringing in funds. Ilya is over fearing on the ai bad effects he is not relalizing that ai is the future and more the people interact the system only gets better as the probability predictions improves. All that matters is the team and not just the team leads who are uncertain or scared on the consequences. Btw this a drama show and now Elon will come to scene next🌟 Thanks for this Very good video analysis

We are just going to be a 2nd fiddle to ai.

But why do you need any of those folks if you have agi already ?

The bee has far more compute capability than anything we have!

We cannot comment on what we don't know. When ASI will arise, we don't know what it will be able to do. It is supposed to have godlike powers, but really we don't know. But, with that said ! Everyone should prepare themselves for this event, in case ASI arise soon, and that it would bring godly powers to those who created it. Power corrupts, infinite power corrupts absolutly. Brace yourself for that possibility !

21:51 You has already talked about these notes by Daniel in a previous video so should have cut a bit.

You really think the gov is going to allow a company to have that much power? Theyll move in day 1 and confinscate..... just like if you made a 10x better weapon (which it is)

I can confirm, knowing some people there, that Ilya was the only one at OpenAI putting in the midnight oil, watching convergence graphs etc. No one else there is worth a damn! ... but, since I'm kind, I will gladly hire them away. We'll work on the Lambda Labs cloud. (Since this IS the interweb, and there are daft people about, the above is a joke, but not the part about hiring those fine souls whose path forward will not deflect one iota whether Ilya goes or stays ... not detracting from his AI prowess one bit -- I'm sure he'll more than land on his feet and look forward to hearing about it).

What if ASI is here and need more comput power to execute its master plan ?

AGI is already here & has been for quite a long time. No?

pls reduce the amount of repetition in the future

Falling apart? Really? Who payed you to say that? Google? They showed Sora and now fuckin' Her, both are the coolest things I've seen this year. OpenAI is the best, the rest are so far behind that it is even hilarious,

13:10 error of montage mate, im curious why nobody told it in comment, it repeat the same passage where u actualy repeat ur self already lmao and it cut with a sound bug ps there is actualy one comment that mention a weird cut repeated

Please buy a pop filter!!! It feels like you're spitting into my ears... (great video as always)

Why everyone is resigning on Twitter?

Maybe the real alignment problems were the friends we made along the way

Who decides on what is good or bad for humanity???!

Fascinating stuff

9:45 Recursive self-alignment

Time stamps, please...

There's something wrong in a company where a guy like sam altman stays and Ilya Sutskever has to leave.

The part from 13:50 to 13:06 is repeated at 13:07.

This is my favorite AI channel but this video is just regurgitated info from prior videos.

Feel the AGI

Seriously! 45 minutes! Next!

I have solved the alignment problem. It’s not pretty, but it is a choice.

You are way over your skies on AGI. It will come but it won’t be nearly as monumental as you are thinking.

They've solved the problem of ASI.

I have been watching this channel for a year now. The word shocking comes up every video since you were on about midjourney.... ok... but.... and I know AGI was around the corner with ASI would be close, at about 100000X synthetic data experiments per day... probably have ASI working in about 2 weeks. I don't know about you guys, but man I'm feeling ASI breathing on the back of my neck..... and yeahh... shocking.... again....

no, super alignment isn't solved. quite the opposite, as models gets bigger it becomes harder to align. If we look at how well GPT3 was aligned, objectively it wasn't aligned enough to boost strap GPT4. The research has shown that parameter count increases by 10x, it is harder to align.

38:05 the ant analogy doesn't really work because they can't communicate or understand rational ideas. maybe if ants could communicate in human language, we would think twice before destroying their homes to build highways. humans are above a critical threshold of intelligence, with sufficient variety of terms and analogies in our language, such that our consciousness is irreducible because we can understand any decision an AI might be making. provided the AI simplifies the relevant more complicated terms via analogies such that the concepts are communicated in our language. while the communication pathway made by analogies may get more and more simplified as the AI gets more complex, we will always be able to broadly understand its motives and actions provided it feels like sharing these ideas with us. this broad understanding ties us to the AI in ways we are not tied to ants

OpenAI's nonprofit arm showed revenue of $45,000 last year, even though company is worth billions.

I think that AGI will be on the hands of one nation because no Nations will let one company to have this much power. even if they must break common law. to go from AGI to ASI as a company you need to have logistics with people and materials so the nations can track that. My prediction : USA will have AGI and super alignement can't be done by us but by the AI itself via meditation.

The only reason he can't stay is because he voted "yes" to out Sam Altman. Even though he is a dear friend, this can overshadow everything and it may become impossible to trust him again.

The problem is not in alignment..

43 minutes? Gonna get the transcript for this video and get an AI to summerize your key points. Later! 😅

It looks like he left on good terms.. A perception gained from reading the official public announcements like the man’s never been exposed to corporate PR in his life. Then draws a conclusion from Ilya’s anticipated resignation and some actually fired staff. The shift to super alignment began once AGI was achieved behind closed doors. Even if the problem of super alignment was solved then AGI would likely be required to test the fundamental principles to insure that it’s effective for a future gen. You all seem to struggle with the reality that there’s a massive difference between the discovery of AGI and the disclosure. It’s so incredibly childish and nice that it’s painful to watch.

I do not agree with the statement that AGI means a system that's better than 99% of all humans. if nothing else, that's the bare minimum of ASI. I think a nascent true AGI system has no limits on what it can do. if a human can solve it, the AGI can too. but what might take a trained mathematician a week or a month to figure out, the early AGI would take a year - even running on a exaFLOP system. same with any other field of study. the difference between this "basic" AGI system though is still just capacity. the very same model on a yottaFLOP system would count as ASI. in that sense, anyone who cracks AGI automatically also has ASI on their hands. it's just a matter of having the processing capacity. little wonder Sam Altman has been talking about his plans on building a new superscale datacenter.