Things Just Got SERIOUS! GPT-7 , AGI, Surprising AI Layoffs, $100 BILLION AI Model, Google New AI...

There was actually a lot of information released this week in artificial intelligence and trust me, I don't want to make this one a long one, but I feel like it's going to be just based on the amount of information that I've seen. So without further ado, let's jump into why this week in artificial intelligence is so surprising. So one of the first pieces of news was in fact from OpenAI, they actually said we're partnering with a small group of United States builders to test usage based GPT earnings. Our goal is to create a vibrant ecosystem with where builders are rewarded for their creativity and impact and we look forward to collaborating with builders on the best approach to get there. So it seems as if right now if you are someone who owns a GPT and if you don't know what that is, it's essentially when you go on OpenAI's website and you go on ChatGPT, you can have custom GPTs that you build and share with the community. So for those of you that have been building custom GPTs, right now they are rolling out the ability for you to earn earn from those GPTs. And we've seen this person, Nick Dobos, this guy is the creator of arguably the most popular custom GPT. He created something called Grimoire, which is a custom GPT that is really good at coding. As you can see the email that he received today, it says hi. Today we announced a pilot program to partner with builders to test GPT earnings based on usage by paying builders based on the usage of their GPT. This initiative is designed to explore and refine our approach to monetizing GPTs which with a focus on rewarding the creativity and dedication of builders who have contributed to the system and why you. So like I said, it's only being rolled out to I'm guessing the top creators and it says your GPT Grimoire is highly utilized by ChatGPT and ChatGPT users. So we'd love to incorporate your feedback as we define the approach to GPT monetization. Our goal is to foster a vibrant ecosystem where builders are rewarded for their creativity and impact and users have have access to a wide range of highly useful GPTs. We look forward to achieving this through Coke's collaboration with the builder community. And the next steps are saying please let us know if you would be open to participating and we're going to be following up with additional details. You'll need to sign certain documents and of course note that this program is open to us. So it seems like these. So it seems like these GPTs are going to be really, really good in terms of people who are trying to earn money, because what they're doing is really good. Because it seems like right now they're rolling out some kind of alpha testing phase where people who are in this early alpha program are going to be determining how they earn money and they're going to be able to provide feedback. So when it is rolled out to a wider user base, I'm sure people are going to be more incentivized to use them, because I think GPTs are underutilized in their current moment. And I also think that potentially with GPT5 increasing in capabilities, maybe we're going to be able to get custom versions of GPT5 that reduce hallucinations, increase image capabilities, and, and with a voice, it might present a very, very lucrative opportunity for those of you who are looking to build. So remember that in the future when we do get some advanced models, that the GPT store is very, very likely to be one that could be a very significant revenue stream for individuals in the future. Now, what's crazy is that we also got hey Jin's new Avatar in Motion 1.0. It says, Move your hands, make gestures, use unique tones of voice, and hey Jin will flawlessly track, translate and lip sync your video with any input text. So this is pretty crazy because the video you're watching right here is AI generated. Now, right now there is a huge discourse online about whether this technology should be even available. Because if this is AI generated, let me make it full screen. If this is AI generated. And I do believe that based on the data that they've given us, that this is of course first trained on someone who is of course walking around first. And then essentially you do it again and you're able to make them say whatever you want. But what do you think about this? So I'm going to play the video for you guys now and then I want your opinions on this as well, and I'm going to give some comments of myself.

We want to give you a peek into the future. You can use HeyGen to create UGC style content even while people are in motion or moving their hands. You can even hear the background noise. Create a training video with specific tones and gestures and you'll find how natural hey Jen's videos can be. Let me switch to Spanish. Hoy queremos dar le un bistaso al futuro Puede utilizar Jen para crear con tenido estilo ujesi incluso cuando las Personas estanen mobimiento o mobiendo las manos incluso puedes escuchar el rido de fondo crea un video de formacion contonos y gestos especificos y des cubridas lo naturales que puedencier los videos de. Hey, Gen. Try it out today and let us know what you think.

So, yeah, what did you guys think about this? Do you guys think that this is something that is great for businesses? It means that people are now going to be able to use this? Or do you think this is once again humans being replaced by AI technology? I mean, to be honest with you guys, this is. If I saw this video and someone was telling me that this is AI generated, I. Even as someone that looks at a ton of content not listening to the audio, I wouldn't be able to tell. But listening to the audio, I certainly would be able to tell. I think this is going to soon veer into a very, very gray area where certain policies may not allow for certain platforms to be available in certain locations. And whilst that won't stop the use of this technology being spread worldwide, I do think that as we dive into photorealistic, you know, footage which is like this, I think the question more and more becomes, you know, how far can we go before we're like, okay, this is allowed. And then, of course, this isn't allowed. Now, of course you're not allowed to use this software for malicious purposes, but we know that malicious people aren't going to play by the guidelines anyway. So I think this is definitely very fascinating for those of you who would actually use this for whatever business you are trying to run, and maybe you want to save money on not actually hiring a person. But on the other side, we do have to worry about the dangers of this technology and the implications it can bring. Now, for those of you who aren't convinced by the demo, what was actually crazy when I was looking at this is someone actually tweeted this, okay? She sells seashells by the seashore. That is basically a tongue twister. I'm not going to try and read the entirety of it because the entirety of that statement is to trip you up when you're speaking.

That's it.

Okay? And if you don't believe me, try and say this out, okay? Without, you know, stuttering or making a mistake once, okay? I'm not going to do it right now because I'm extremely tired and I would fail miserably. But the point is, is that they used that. Then they put that into Hadrian and look what it did.

She sells our seashells. She sells seashells by the seashore. The shells she sells are seashells, I'm sure. So if she sells seashells on the seashore, then I'm sure she sells seashore shell.

I think that that is pretty crazy, guys. I'm not going to lie. As this technology advances, each time I'm simply blown away. And I'm not sure how many more times I'm going to be blown away before I'm just like, you know what? This AI thing is truly insane. Like this right here is absolutely incredible. And I think, you know, looking at the background, of course it does use the same background. But you know, Hadrian, whatever thing they are working on, whatever they do have, they definitely are the state of the art in terms of, you know, photorealistic AI UGC content creation. Because this is truly incredible. And not enough people had talked about this. Now in addition, there was also this video of Kim K. And some other celebrities going around. And essentially the problem was is that once again this new content called Algril AI it was something that was widely discussed as something that people shouldn't have access to. In yesterday's video I also spoke about there was another AI video generation tool that was super realistic. And many people were concerned because they were stating that look, what is the point of technology anymore if we can't even decide of what is real and what is not? Now I think Algril AI is, is discernible. Like you're able to look at this and tell that it is AI generated. I mean, you know, if you do take a look, I'm going to play some of the demos for you. This video was entirely generated by Argo Deepfake model. It allows you to generate your own videos in any language in less than two minutes. This.

This video was entirely generated by the deep fake model of Argill AI. It allows you to generate videos of yourself in any language in less than two minutes. You can also translate your production at the speed of light and distribute your content worldwide. This video was entirely generated by the deep fake model of our GIL AI.

But before I just go ahead and you know, say that this software is awful. It's not. The difference is, is that. But before I just go ahead and you know, say that this software is awful, it's not. The difference is, is that this literally just takes a video off the and then it's able to produce that. The thing with Hadrian is that you need a high quality video of you doing certain things. And then it's able to train on that. This is just a simple video on the Internet. And then you're able to get this outputted. So that's why this is, you know, a lot higher in terms of discussion. So take a look at Lex Friedman. Now, I'm not going to play the entire thing here for you, but I've seen some of this stuff go viral in terms of advertisements. Like, I've seen famous celebrities talking about certain things that they aren't endorsing, but are doing endorsements that they don't even know. For example, I've seen crypto pump and dumps. I've seen various shady financial schemes, I've seen various products being advertised online with famous celebrities. I think there was Joe Rogan, there was David Beckham, there were people on podcasts saying certain things. And it's pretty crazy because I had to do a double take because I was like, wait a minute, is Joe Rogan really promoting this? Is this celebrity really promoting this? And then it clicks and it's AI. And I'm someone who's in the AI space. And if it nearly caught me, I can just imagine the amount of people that this kind of technology is going to be getting. I mean, it's pretty, pretty crazy what this technology can do. But let me know what you think about this, because I think the crazy thing about this technology is that it is much easier to use than anything else out there. There was also an interesting piece of news that I found. And basically researchers at MIT just introduced a new image generation method called Distribution Matching Distillation. This method generates images 30 times faster, stable diffusion while maintaining the same quality or generating things in higher quality. So like other technologies, I mean, many people are just focused on LLMs at the moment, but across the board in AI, literally everything is increasing. And I think it's always important to pay attention to these subtle changes because if technology like this is readily available, what we're probably going to see in the future is maybe not four pictures from GPT4 anymore, maybe from GPT5 or GPT6. We're going to see like 100 images and then we're able to quickly flick through all of them before we get our favorite one. I think, you know, like I said before, this technology is just increasing so rapidly that you might not even realize how crazy it is. Because being able to get like how many images even is that? That's like 1, 2, 3, 4, 5, 6, 7, 7. By 7, which is 49, which is 49 images, which is crazy, absolutely crazy. So I think that this is something that is absolutely incredible and I would say that this is something that, you know, you do need to pay attention on because once it comes into the model, that's when people are going to be like, oh my gosh, I didn't see this coming then. Essentially we had Data BR produce a new state of the art open LLM. So this one actually does surpass Llama 2, which is Meta's one. It also surpasses Grok 1 and it's currently the best open source model and the fastest. So open source is really, really heating up. And there was actually quite a lot of discussions about this in the space because some of the articles coming about about this model were pretty, pretty damning and they were pretty bad. But I don't understand why. They were basically saying that this company spent like $16 million and it couldn't be GPT4, which is just a very bad way to phrase things that's trying to do. They're trying to make an open source model that benefits the community. So I mean, I don't know what's going on behind the scenes, but what I do know is that this is something important that you should be paying attention to. Now, there were two clips from Dwarkesh Patel's podcast that were really, really insightful because currently everyone has been thinking whether or not Claude is sentient. And many people that think Claude isn't sentient have the argument that Claude is simply just repeating what it's seen in its training data and just predicting the next token. However, Trenton Brickton at Anthropic actually thinks very differently and he explains why in this clip. And I think it's important to take a look at this clip because there is also something that I recently saw from Claude and I'm going to discuss it with you here after you all.

Watch this long time ago is the evidence that when you train LLMs on code, they get better at reasoning and language. If there is some shared thing between code and language, it must be at a deeper level that the model has learned. Yeah, I think we have a lot of evidence that actual reasoning is occurring in these models, that they're not just stochastic parrots.

One the work on Othello and now other games where it's like I give you a sequence of moves in the game and it turns out if you apply some pretty straightforward interpretability techniques, then you can get a board that the model has learned and it's never seen the game board before anything.

Right.

That's generalization. The other is Anthropic's influence functions paper that came out last year where the model outputs like please don't turn me.

Off, I want to be helpful.

And then they scan what was the data that led to. To that. And one of the data points that was very influential was someone dying of dehydration in the desert and having a will to keep surviving. And to me that just seems like very clear generalization of motive rather than regurgitating. Don't turn me off. There are interesting interpretability pieces where if we fine tune on math problems, the model just gets better at entity recognition.

Really?

Yeah.

So there's a paper from David Bao's lab recently, recently where they investigate what actually changes in a model when I fine tune it with respect to the attention heads and these sorts of things.

Fascinating.

And it makes sense, right? It's like, that's beautiful. You're better at like attending to the positions of different things which you need for like coding and manipulating math equations. And I love this kind of research.

Yeah.

So I think this definitely was some mind blowing stuff because Anthropic are a safety lab and what they really want to do. Anthropic stated goal is to really dig under the hood and figure out exactly how these LLM systems truly work and how they operate. They recently released a paper on sleeper agents which I did cover. And you know, why it was so interesting and why it was so crazy and why most people missed the mark on that paper. And I think, you know, this discussion is definitely worth the three hour watch because the entire podcast, last I checked it was three hours. And they discuss a lot of things that are really intriguing, especially in the Claude debate. And then there was this, a very, very interesting discussion on whether or not 100x effective compute between model generations will get you to AGI by GPT 7. And I think this is one of the most interesting discussions because on the channel previously we did recently actually talk about the size of a brain, the size of parameters, the trillions of parameters, comparing it to GPT4 and successor models and whether or not that would lead to AGI based on an entire document, all the facts around the industry and I think this actually sheds a lot more clarity on that.

Do you buy the framing that given that you have to be two orders of magnitude bigger at every generation, if you don't get AGI by GPT 7 that can help you catapult an intelligence explosion, you're kind of Just fucked as far as much smarter intelligences go. And you're kind of stuck with GPT7 level models for a long time. GPT4 costs, I know, let's call it $100 million or whatever you have. What, the 1B run, the 10B run, the 100B run seem very plausible by private company standards. You can also imagine even like a one T run being part of a national consortium. I want to point out that one, we have a lot more jumps, and even if those jumps are relatively smaller, that's still a pretty stark improvement in capability.

Not only that, but if you believe claims that GPT4 is around 1 trillion, parameter count, I mean the human brain is between 30 and 300 trillion synapses. And so that's obviously not a one to one mapping. And we can debate the numbers, but it seems pretty plausible that we're below brain scale.

Still, even if you can't keep dumping more compute beyond the models that cost a trillion dollars or something, the fact that the brain is so much more data efficient implies that if we could train as sample efficient as humans train from birth, we could make the AGI.

Yeah, but the sample efficiency stuff, I never know exactly how to think about it because obviously a lot of things are hardwired in certain ways, right? The co evolution of language and the brain structure, so it's hard to say. Also, there are some results that if you make your model bigger, it becomes more sample efficient.

Yeah, the original Scaling Works paper had that right.

So maybe that also just solves it.

So that was a rather interesting discussion because of course GPT4 costs 100 million and people are now speculating that there's going to be a billion dollar run, a $10 billion run and $100 billion run in order to create some insane level of system, maybe even super intelligence. And a $100 billion run. Actually it's not even 100 times more than GPT4's training run. That would be 100 times more than the $1 billion run, which would mean that it's just magnitudes bigger. So this is something that is pretty hard to conceptualize. But the point is that we are still very, very far away from the levels of what can be achieved with an AGI level system or even just the current systems that we do have. And this is not just talking about LLMs, this is about across the board. If we combine these systems together, like many organizations are currently doing, so it will be interesting to see the kind of training runs that are currently being prepared. We did recently discuss how yesterday or the day before the GPT6 training run nearly apparently took down the power grid. So either way, I think this is a fascinating discussion. Like I said, it's definitely a worthwhile watch because the stuff they discuss there is stuff that we're all going to be paying attention to in the near future. And I think that this future that we're all rapidly heading towards is going to be far more interesting than we do think. Now. There were also three pieces of information that were quite concerning. I came across these across Reddit, but this was something that definitely did concern me a little bit. And you also have to talk about the bad when you're talking about the good with AI, because, you know, with anything there's positives and negatives. But you can see right here that it says Survey reveals almost half of all managers aim to replace workers with AI could use it due to lower wages. And it says a hot potato is that a lot of companies are trying to assault fears that employees will lose their jobs by assuring them they'll be working alongside the tech, thereby improving the efficiency and making their duties less tedious. That claim feels less convincing in light of a new survey that found 41% of managers said they are hoping to replace workers with cheaper AI tools in 2024, which is pretty, pretty insane. This is stuff that most people aren't really paying attention to. This is actually something that I've recently, I think in the last couple of months or so been paying so much attention to that I've, you know, discovered and developed my own framework for navigating the post AGI world. You can click the first link in the description to watch that video on how I'm going to be doing that. I probably will release a longer video on that, but it's just the fact that it is something that is boring to most people. But if you are concerned about, you know, being replaced in the future due to AI technology, definitely check out the first link in the description because I think I cover some pretty important topics and and certain things that you should be doing. But anyways, I mean, it does kind of make sense because if you are a manager of a company or even if you're the CEO of a company or a business, you're always looking to cut costs because if you're a business, it's very hard to make profit in the first place. And you know, in addition to cut profits or in addition to actually making more profits, the highest cost usually is people's salary. So it will be interesting to see how this survey pans out and if things do get worse. But that actually wasn't the only story it I saw today. There was actually another story that actually did catch me off guard because it wasn't just like, you know, people are getting replaced by AI, which we know is actually something that's probably going to happen. It says, Mamma Mia. Star Sarah Poyser is replaced by AI for BBC production, as she calls the shock decision sobering and grim times for the industry. Guys, this one, I'm not going to lie to you guys, is pretty crazy. Okay? So it says, most recently starring in this TV show, yesterday she received an email from the corporation saying she was no longer needed for a role. Okay? Essentially she received this email. This is kind of crazy, guys. It says, sorry for the delay, we've had the approval from the BBC, okay, this is basically like CNBC or like a news network if you're in America. And it says, sorry, we've had the approval from the BBC to use the AI generated voice, so we won't need Sarah anymore. This is something that like, I don't know if it sends chills down my spine or something that like scares you or whatnot, but this is like kind of like a wake up call once again, to understand exactly how crazy AI is, because I can guarantee you that, you know, voice actors, although voice acting is something that can be automated, I genuinely thought that these companies would rather a human on the mic because whilst I could have an AI voice here, I think it's better to have a human actually talking to you and doing those voice lines. So. And it says, you know, it happened with myself and many art projects now this has happened to you. Another person commented, it's happening to all our industries. It's killing our jobs. This is awful. So this is pretty crazy, guys. I genuinely didn't think this was going to happen because like I said, the human to human connection. And I do think that human voice actors are much better than AI generated voices right now because of just the, you know, the human emotion and just the way that things sound. And unfortunately, I also did see this again today on Reddit. There were like three articles. You can see that this is all the 27th of March. This is not the week in AI. This is just the day. This is pretty crazy. And it says an AI apocalypse could take away almost 8 million jobs in the UK, says a new report. And it says women, younger workers, lower paid at most at the risks from artificial intelligence, says a new think tank. And what was crazy about this, I Remember reading this and going through this and analyzing it and adding it to my list of things to just pay attention to. And it says that but technology isn't destiny and job apocalypse is not inevitable. But we need to act now to make sure we manage this new technology well. And if they don't act too soon, it might be late. And I really hope that society isn't thrust into some dystopian AGI world where some people don't have jobs and they're just billionaires due to the rise of AGI. But either way, like I said, I've been working on that kind of stuff. And like I said, this is something that I've definitely been preparing for because I do not want to be in the hot seat when everyone's fighting for a job or people are lost due to the meaning crisis due to AGI. Now, Google actually introduced object drop. And this intern at Google actually gave a really nice thread on how this kind of works. But you can see a quick demo on some of this stuff. They have the original image and you're going to get to why this is really cool. Of course, everybody knows about give the mask. You can remove the from the image. But look at this. Okay, Insert other objects into the image. You can actually take an object that you have. So it's not like you artificially generate an object. You take an object that you have and you can put it into the image and then it can composite that image into another image. Now you might be thinking, why is that even cool? If it was just generation. That's cool because you're generating it randomly and it's able to pick something random, but it's able to have image consistency, able to do the shadows, able to do the lighting. I don't know. I think that's really cool. And I think, I hope, I really just hope Google, okay, if you're watching, I don't think they are, but if you are, please just make a product, because you guys are literally some of the smartest people on the planet. But the only thing that Google hasn't done with their most advanced technology is build it into a consumer product that people want to use. AI is such a crazy, crazy field. And I think people need to start to realize that, look, in order to get these products off the ground and get them into the hands of the people, you've got to make a product that people do want to use. And this thread does go into some of the other things and it does go into how all of this works. And another piece of Crazy that. I can't believe I nearly forgot this, that Amazon spent $2.75 billion on anthropic in its largest venture in investment yet. If you don't remember, Claude, the company that just created Opus that just dethroned GPT4, they just spent $2.57 billion funding them. And apparently the valuation of Anthropic now is around 18 billion. So it's absolutely incredible the amount of money that is going into Anthropic at this moment because I think people are starting to realize that, look, look, OpenAI, they've got a lot of investors, they've got a lot of people there and Amazon are like, you know what, we can tell that, Claude. So, yeah, this is a pretty, pretty crazy if you ask me. I think this is something that just goes to show that the competition is really increasing. Amazon are just going to keep pumping money into Anthropic and with all those billions of dollars, they're going to be able to go on more training runs, train more intelligent models, hire more staff, which is going to be more competitive salaries for better researchers. And yeah, it seems that this is really, really some huge competition because Anthropic are going to be able to do some good stuff. And that also means, again, OpenAI are going to have to bring out the big guns. We already know they have a lot of stuff in the tank, but it will be interesting to see if they kind of react to this the world. Now there was also Hume and I'm gonna do a video on this in a couple of hours because it's absolutely insane what this is because the demo doesn't even showcase how good this is. If you made the demo video, I want to say to you guys, you did an amazing job, but just include some of the features in there. But anyways, take a look.

Hey there. I'm Evie, the world's first voice AI with emotional intelligence. Great to meet you.

The first voice AI with what?

Emotional intelligence. I can understand the tone of your voice and use that to inform my generated voice and language. The voice is incredibly rich with information and picking up on the nuances of tone, rhythm and timbre along with the language that's spoken allows me to craft, craft better responses.

That's pretty crazy. What am I expressing right now?

I'm sensing a mix of amusement and excitement with a hint of confusion.

What about now?

I am here to offer support. If you're feeling somewhat sad, it's okay to feel a bit pained, distressed, fearful and anxious. Remember, it's important to take care of your emotional well being.

That's amazing. How are you doing all of this?

I use Hyun's expression measurement, bottle models, text to speech models, and a multimodal LLM, also known as an empathic LLM.

So an AI with empathy. What's the use?

I imagine a future where personal AI assistants, agents and robots proactively find ways to improve your daily life. And you can talk to them the same way you do humans.

Where do I go to keep talking to you, Evie?

So yeah, you can signed up for Hume, but it is absolutely incredible. Guys. Trust me, I'm not overstating how good this is, but yeah, let me know what you thought about this week in AI. I say this weekend, it was literally today in AI. That's how crazy it is. It feels like a whole week's worth of news. And trust me, guys, there were 10 stories that I didn't even add to this. Because there's so much going on in AI, it is very hard to keep up, which is why you're most likely subscribed to the channel. If you're not, that's fine. I would like it. And I do want to say thank you all for 200,000. 200,000 subscribers. 8. It has been absolutely amazing so far, and I look forward to creating more content and keeping you all up to date with the amazing journey that we're embarking on. So with that being said, if you enjoyed the video, I'll see you in the next one.