OpenAI Employees FIRED, 10X Compute IN AI, Agi Levels, NEW AI Music , Infinite Context Length

Okay, so there was some actual interesting news in regards to OpenAI. And one of the things that I found out was that some OpenAI researchers, including an ally of Sutkova, were fired for alleged leaking. This is pretty, pretty crazy news. We can see here that OpenAI has fired two researchers for allegedly leaking information. According to a person with knowledge of the situation. They include Leopold Aschenbrenner, a researcher on the team dedicated to keeping artificial intelligence safe for society. Society. Ashton Brenner was also an ally of OpenAI chief scientist Ilya Sutskeva, who participated in a failed effort to force out CEO Sam Altman last fall. It's not clear what information the two staffers leaked. The other staffer, Pavel Ismailov, a researcher who worked on Reasoning, had also spent time on the Safety team. And the ouster of the two men is among the first staffing changes that have surfaced since publicly. Sam Altman resumed his board seat in March, and his. It's pretty, pretty crazy because these two were on some of the, you know, Super Alignment research papers that were really, really important. And it, you know, says that here says internally, Ashen Brenner was one of the faces of what OpenAI calls its super Alignment team. And that was the team in which they were trying. Well, not were. They are, you know, pretty much trying to solve Super Alignment with, you know, artificial super intelligent systems, if that makes any sense. But asi. So they're trying to align, you know, those really powerful systems, and they were doing that with Sutskova. So it's pretty interesting that, you know, many people were stating that. Why did they even mention that these guys were allies of Satska? I think it's kind of interesting because, you know, Sutskova's role at the company is unknown. There have been, you know, numerous rumors circulating around, wondering where he is, what he's doing. And then, of course, two people who are quote, unquote, allies with Satskova, who were also working on Super Alignment, are now fired. So, I don't know, it's kind of interesting to just, you know, see what's going on at OpenAI. And you can see that it says Satska, co founder responsible for OpenAI's biggest breakthrough, was part of the board that fired Sam Altman. And, yeah, he's been largely absent from OpenAI since the absent. Brenner, when reached by phone, did not have an immediate comment and Ismailov did not have a request to comment. So, yeah, it's pretty crazy. And the lawyer who has represented Suskavar did not represent, you know, for for a comment. So, I mean, the situation here is pretty interesting. A lot of people are now speculating, wondering whether that means OpenAI is going to get any more leaks. But I don't know what kind of leaks they were doing. Some people have speculated that they were leaking to, you know, foreign intelligence or foreign nations or whatever. That's just pure speculation. Some people are saying that, you know, it's links to Jimmy Apples and, you know, the Twitter leaks. But I mean, I gotta be honest, I think whatever leak this is, I do think it has to be really serious for them to get fired over it, because the Jimmy Apple's leaks are like, not substantial information. It's not like it's an entire document that he posts on Twitter, you know, stating what the new models are going to have. It's usually a vague tweet from account that has like three followers. And then long after the, you know, information has been published, he retweets it and says, look, I knew this information is before. And of course he does have inside information, but it's just not usually leaked in the way that it as like, you know, would seem to lead to a firing. Maybe. OpenAI is really, really strict, but I'm guessing that it had to be probably serious for them to fire two members of their really important team. I know that, yes. You know, as well, another thing that I do know is that, you know, AI talent right now is really, really hard to come by in the sense that all these labs are fighting over, you know, people like these two, these individuals. But I wonder how that actually impacts their market value in terms of finding another career. I don't think it's going to be too bad considering how competitive the AI space is in terms of trying to find AI researchers that can, you know, work on reasoning and stuff like that, and especially trying to get to AGI. I'm just thinking that, you know, potentially it may have been, you know, more serious than we think, considering that they just decided to fire them. Either way, it's pure speculation. I don't think we're ever going to know because if it is serious, it's probably going to be lawyers involved, probably going to be stuff that's not disclosed to us. So, yeah, we'll be interesting to see if there is any more information on that. But there was some other interesting news. There was also this. A voice agent indistributable from humans handling interruptions smoothly and responding to the, you know, person. So take a look at this because this was pretty fascinating.

Hi there. This is Mia From Ace Plumbing. How can I help you tonight?

Hey Mia. Yeah, I have an emergency here. We've got water leaking from our upstairs bathroom. It must be a burst pipe.

Oh no, that sounds stressful. I can definitely help with that. We do have an extra service charge for off our emergency services of $150. Is the pipe still leaking?

Yeah, that charge is fine. And yeah, it's still leaking.

Okay, got it. Do you know how to shut off the water main to your house? If not, I can, I know how to do that.

Don't worry about that. I will do that right after this call. Great.

Do you still want a plumber to come tonight? If so, may I have your name and address please?

Yes, we do. It's Kara Smith at 146 Stevens Road in Alamo.

Okay, let me check the schedule. Um, yes, we can have someone at your house at 11pm thank you for providing that information.

Sounds good. Thanks so much.

You're welcome, Kara. Is there anything else I can help with? If not, thank you for calling. If anything else comes up, feel free to give us a call back. Goodbye.

Okay, so I want to ask you guys this. What do you think about this kind of software? Do you think that this is good for reducing overhead in terms of companies or do you think this is just mindless automation that leads to, you know, people not doing the work? I do think that this kind of work isn't exactly particularly fulfilling. So I don't think think it's as bad. But I'm, you know, intrigued to know your opinions because I'm someone that when I call up a support place and then I have to talk to an AI system, it's frustrating, okay? But I think our frustration comes from the fact that we don't understand these AI systems. So in the future, if these AI systems are actually, you know, infinitely smarter and they actually do work as well as a human, will there be as big of a, you know, disagreement with these systems in place? Because that's what I want to ask you guys. And I don't think there will be considering the fact that some people just want their issue resolved. And of course some people like myself, I think by that time we might be considered old fashioned. And I'm guessing that, you know, the newer generation, the generation growing up will probably be just knowing that they've always spoken to AI systems and not potentially people. So maybe it's just going to be a generational thing that changes as the new generations come in. Then we have introducing Yu Gi. Oh, and this was by far One of the biggest announcements that I will cover in a more introduction, detailed video, because there's actually a lot that people are missing. But take a look at this demo video, and then I'm going to show you guys something that I found to be just absolutely hilarious from this.

Celebrating.

So.

Let'S go. Oh, we are so bad. So it's never been more bad than it is right now. So bad. So bad.

Make those ears go down.

Break it down with that rolling sound. So, yeah, that was udio.com. maybe you skipped the demo and all the music of listening to it, but I think it's important to actually look at the prompts. And there was. This is probably the funniest part of the video. And I just think this is honestly so funny. Okay. And apologies to Seth in advance, but I found this tweet, okay. And this is basically on the platform, and someone found an account on the Yu Dio platform that essentially is generating songs about how his friend Seth managed to, you know, I guess you could say, release his bowels unexpectedly at work. And honestly, the fact that it sounds so professional is just. It's just hilarious. Okay, this is. This is the funniest thing I've seen all day. I honestly cannot stop laughing at this. But I'm guessing that some people are just continually to make songs about their friend who managed to release their bowels at work. And the songs are actually just hilarious. Take a listen.

Ted shit his pants, had Gladstones. He was a grown man who pooped in his pants, and Dave had to drive him home.

Oh, man, that. That just never gets.

I got me a man. But. But there's just one problem. Well, he works 9 to 5, and he's a hell of a guy, but he's in his pants at work.

Oh, Seth, I'm sorry, but, like, that's just the funniest thing I've ever seen. And I think. I think the use cases for this are gonna be pretty hilarious. I mean, there's so many custom soundtracks and creative ways that people are gonna use this. Hopefully Seth doesn't mind that, but it's pretty funny. Okay. It's pretty. Pretty funny. Okay. But the point is, is that the quality is absolutely incredible. And if you're using Yu Dio, it's actually been created by former researchers at Google. They made a previous paper called Lyra, and. Oh, no, it was either Lyra or Lyria. To apologies. But the point is, is that Google researchers, I think, are increasingly leaving Google to create products. And I'm wondering if Google is even going to be standing in the future because recently Demos Hassabis talked about starting up his own AI lab and that's the current Google AI CEO. And the reason is Google just isn't putting out the products Google has literally. I've looked through so many Google research papers. When I tell you they have a boatload of stuff that is, I wouldn't say it's ready to release, but they could really, really crush the AI game if they wanted to. But for some reason I don't know what's going on with the governance structure, how things are organized there, but for some reason they, they're taking forever to release products and it's frustrating people, which is why people leave and they create software like this that people can use to make stuff like this. So it will be interesting to see if Google blows its lead because honestly, the way things are shaping up, I think they're gonna. Now we have a clip from Ray Kurzweil, a legendary futurist, talking about how even his technological progress timelines, it's moving faster than his predictions.

If you have been at the center of this extraordinary last few years, can I ask you, is it moving faster than you expected it to? How does it feel to you?

It feels like a few years. I mean I made a prediction in 1999. It feels like we're two or three years ahead of that, so still pretty close.

Jeffrey, how about you?

I think for everybody except Ray, it's moving faster than we expected.

There was also a research paper from Google that was Infinite context Windows, which is, it's pretty insane. I'm not gonna, you know, spend too much time talking about it. But the point is, is that if this application is true, then that means we could get infinite context Windows, which could mean absolutely insane ramifications for the future. And the reason this is so crazy is that I remember when I was talking, I think eight months ago about how 100k context lengths or are just absolutely incredible and now it went from 100k to 128k to 500k to a million and now it's infinite. Like it, like trying to wrap your head around how fast AI is moving is pretty, pretty surprising. So I mean, you know, like right now I can't even speak, probably can't even make a sentence coherently to present this video. But the point is, is, you know, it's, it's just incredible how crazy this is moving. And people like me are even shocked.

So to speak, in the amount of compute used to train the cutting edge AI models Per year. So instead of doubling per year, which is the Moore's Law trend, we're increasing the amount of compute by 10 times per year. Because in this case we don't need the compute to be smaller. We can just daisy chain more computers together. So our server farmer inflection, for example, is the size of four football pitches, right? It's absolutely astronomical, uses like 50 megawatt of power. And you know, so you look at it, it's like absolutely mind blowing. It roars like, like an engine. And all of that is, is really just graphics cards, you know, just like you have, in your, in your, you know, if you have a desktop gaming machine, you might have a GPU graphics card. We just daisy chain tens of thousands of these up together so that they can do parallel processing on, you know, trillions of words from the Open Web. This 10x increase in the amount of compute used to train.

So yeah, it's pretty incredible how much compute is going up in terms of what they're using to train cutting edge AI. And him saying that it's, you know, four football fields is absolutely insane. And we know that OpenAI are building, you know, Stargate, which is going to be an insane level of compute in the future. So I mean, it's just going to be fascinating to see how Moore's Law continues to progress in terms of, you know, AI development because it's going to make things a lot faster, a lot more efficient and of course a lot more crazy in terms of the capabilities that we're going to see in systems because we're going to be able to train models faster and deploy them faster.

Definitely intuition involved there, there was Monte Carlo rollout too. But it's playing with intuition about what moves to consider and how good the position is for us. It's had neural nets for that that capture intuition. And so I see no reason to think it might not be creative. In fact, for the large language models, as Ray pointed out, they know much more than we do and they know it in far fewer connections. We have about 100 trillion synapses, they have about a trillion connections. So what they're doing is they're compressing a huge amount of information into not that many connections. And that means they're very good at seeing the similarities between different things. They have to see the similarities between all sorts of different things to compress the information into their connections. That means they've seen all sorts of analogies that people haven't seen because they know about all sorts of things that no one person knows. About and that's, I think, the source of creativity.

So, yeah, that was a fascinating, fascinating piece of insight from Jeffrey Hinton. Jeffrey Hinton, and he talks about, you know, the previous creative advances that we've seen in systems like AlphaGo. And I think in the future we're probably going to get to a really creative system because, you know, creativity does come from what, you know, and trying to think in creative and abstract ways. And then we're going to be able to get systems which, you know, of course, know more than any human on the planet. And combining that with certain technologies, I mean, we're probably going to get really, really creative systems, especially with hallucinations and all these kind of different things. I mean, it's just an interesting, interesting future to be a part of everyone.

BULLISH LLM and concretely, we believe that when we look at very complex scenes of SMART can generate, like that snowy scene in Tokyo that we saw in the very beginning, that SORA is already beginning to show a detailed understanding of how humans interact with one another, how they have physical contact with one another. And as we continue to scale this paradigm, we think eventually it's going to have to model how humans think. The only way you can generate truly realistic video with truly realistic sequences of actions is if you have an internal model of how all objects, humans, et cetera, environments, work. And so we think this is how SORA is going to contribute to AGI. So I'm going to spend a bit of time talking about that.

One last question.

Thanks for the great talk. So my question is on the training data, so how much training data do you estimate that is required for us to get to AGI? And do you think we have enough data on the Internet? Yeah, that's a good question. I think we have enough data to get to AGI. And I also think people always come up with creative ways to improve things. And when we hit limitations, we find creative ways to improve the pulse regardless. So I think that whatever data we have will be enough to get the H. Wonderful.

Okay, that's to AGI.

Most people, including most scientists, have a particular view of what the mind is that I think is utterly wrong. So they have this inner theater notion. The idea is that what we really see is this inner theater called our mind. And so, for example, if I tell you I have the subjective experience of little pink elephants floating in front of me, most people interpret that as there's some inner theater. And in this inner theater that only I can see, there's little pink elephants. And if you ask what they're made of, philosophers will tell you they're made of qualia. And I think that whole view is complete nonsense. And we're not going to be able to understand whether these things are sentient until we get over this ridiculous view of what the mind is. So let me give you an alternative view. And once I've given you this alternative view, I'm going to try and convince you that chatbots are already sentient. But I didn't want to use the word sentience. I want to talk about subjective experience. It's just a bit less controversial because it doesn't have the kind of self reflexive aspect of consciousness. So if we analyze what it means when I say I see little pink elephants floating in front of me, what's really going on is I'm trying to tell you what my perceptual system is telling me when my perceptual system's going wrong. And it wouldn't be any use for me to tell you which neurons are firing. But what I can tell you is what would have to be out there in the world for my perceptual system to be working correctly. And so when I say I see little pink elephants floating in front of me, you can translate that into, if there were little pink elephants out there in the world, my perceptual system would be working properly. And notice the last thing I said didn't contain the phrase subjective experience, but it explains what a subjective experience is. It's a hypothetical state of the world that allows me to convey to you what my perceptual system's telling me.

So the last clip of Geoffrey Hinton talking about AI chatbots have sentient and subjective experience because there's no such thing as Qualia and Agent House talking about the future of AGI. Let me know what your most interesting thing was this week.