AI Entrepreneur Matthew Berman On The Power of LLMs - "It blows my mind."

In the long run, who wins?

Oh, man. Let's just say I'm a big believer in open source.

I hope you're right.

Yeah, they're taking the scorched earth mentality, the scorched earth strategy. And by the way, I'm a hyper competitive person, so I love it.

Hey, welcome to the Next Wave podcast. My name is Matt Wolfe. I'm here with my co host Nathan Lands, and today we're talking to a serial entrepreneur and AI expert, Matthew Berman. He's got a very popular AI YouTube channel, but let's just go ahead and get right into it. I'm curious about your story a little bit. I, you know, we see all of your YouTube videos. You seem to be super tapped into the AI world, but how did you get into AI in the first place?

I saw ChatGPT along with the rest of the world and I was completely enamored with it. So I decided to start a YouTube channel just to document my learning process and hopefully share my learnings with other people. And luckily my third video went pretty viral. It was about Leonardo AI and I thought, oh well, this is easy. And yeah, the channel just went from there. And midway through last year, I went full time on it.

Being a know it all used to be considered a bad thing, but in business it's everything. Because right now most businesses only use 20% of their data. Unless you have HubSpot, where data that's buried in emails, call logs and meeting notes become insights that help you grow your business. Because when you know more, you grow more. Visit HubSpot.com to learn more.

When it comes to YouTube and creating a lot of this AI content, like, what's your source of information for all of this? It's one of the questions I know I get asked a lot is like, how do you keep your finger on the pulse of all this stuff? And I know for me, just keeping my finger on the pulse is literally my full time job now. So it seems like you've sort of found like this niche for yourself. Like I mentioned you in my videos and I know a lot of the other AI creators mention you in their videos now as like the guy to go watch whenever a new large language model comes out. Right? Like Phi 3 just came out, right? And I was like, Phi 3's out. I could talk about it, but Matthew Berman's going to do a better job than me. So go check out his channel because he'll probably make a video about it. So, like, I love that you're doing that. How did you Sort of fall into that. Like what is it about the like large language models and sort of comparing them and figuring them out that made you want to go down that rabbit hole.

I had this real love for the local model, being able to download a model, have it on my computer, run it, and I know this isn't exactly true, but essentially have the entirety of world knowledge in like just a handful of gigabytes on your computer. Still just saying it out loud blows my mind. And so I wanted to benchmark the models that I was playing with.

I'm curious right now, what's the best local model you could try right now for tech?

Yeah, I'm going to mention two companies and I'm sure everybody watching this video has heard of them. Obviously meta meta AIs with the llama 3 models, those are probably the best and maybe only slightly better than the Mistral companies model. So Mistral AI has Mistral, they have Mixtral which is a mixture of experts model. If I am going to recommend a local model, obviously it depends on how much RAM you have, how much VRAM you have, but generally you're going to choose either one of the Mistral models or one of the Llama models. Now Microsoft just recently released the Phi model, Phi Phi 3 and those are pretty capable. They tend to be smaller and they tend to need more fine tuning for specific tasks. So they don't have as broad of knowledge but they're really performant and they are still very high quality. So I think between those three families of models, probably my favorite is going to be the Llama 3 models. But all three of them are fantastic.

Yeah, yeah, no, I've used Llama 3 quite a bit. Especially when you combine it with Grok. Not Grok Gro but Groq Grok. When you combine it with that version of, with the, what are they called? Is it the lpu, their language processing unit or, or something. But when you use that it's just like insane. Like the speed that you get back from the models on there, the Grok.

Co. Groq and it's kind of nuts. They, you know llama 3 like 8B or it's like 800 tokens per second and you would think like okay, wow, you achieved something incredible. Now like take a break. But no, no, now they're like okay, that wasn't enough. We're just going to continue to work on increasing speeds. And so it's. I can't remember the exact number but they have some absolutely insane input token speeds. I think it's like 2,000 tokens per second, don't quote me on that. But yeah, that company is kind of nuts. They are all about the speed and they're doing really well.

So would you say they're like a competitor to Nvidia? Are they making chips to compete with Nvidia or do they sort of work in tandem with Nvidia chips?

So they are definitely a competitor to Nvidia. However, Nvidia doesn't actually offer inference as a service. They sell you the chips, right. And then like these big data center companies can buy the chips and offer inference. Grok Groq, they used to sell chips and offer inference and they actually acquired this company, I'm forgetting the name of it now, but they essentially acquired a company that's like the front end of the inference service. And so like I mentioned, they used to sell chips and then at a certain point, maybe a month ago they decided to stop selling chips and, and just offer inference. So they are building out massive data centers with their own chip technology and just strictly offering an endpoint, an API endpoint or a cloud service like Chad, except it's lightning fast and you can use different open source models.

So are they trying to be like cheaper, faster, but lower quality, but in some use cases that's okay? Is that the kind of.

Yeah, you're probably right, Nathan. It's like if you're, if you're talking about like that last 5% of quality GPT4O, like it's going to win. But you know, open source is catching.

Up quickly when it comes to large language models. Have you found that like this model's good for X and this model's good for Y and this model's good for Z. Right. Like have you found certain models to be better for specific use cases?

Yeah, and actually that's really why I like open source because these companies like Meta will put out this raw model like llama3 and then Eric Hartford will make the dolphin flavor of it. And all of these versions are very good at a particular thing, whether it's being uncensored or whether it's role playing, math coding. In fact, Mistral just released Codestrol, I believe it's called. It was just today, so I haven't had a chance to look at it. But yeah, so that is also why I love open source because you can have all of these fine tuned models specific to different use cases, where in that particular use case, nine times out of 10 they'll be better than a GPT4 more broadly speaking. Right. So GPT4 is like across the board. GPT4O is just a better model. But if you look at each individual use case and you find the best open source use case, typically you can find one that is just as good and yes, much cheaper and you can get it to be much faster.

It's interesting because you've got OpenAI, you've got anthropic, they're both sort of closed models, right. You've got to basically use them through their website. You can't install them locally. We don't really know what they're trained on. It's kind of all closed off. And then you've got the open models like Llama, Mixtral 8x7B.

Google has Gemini, that's closed source and Gemma, that's open source.

Right. And then Mistral also has a mix of both closed and open source stuff as well.

Exactly.

In the long run, who wins, do you think? I know that's a very loaded question, but like, I'm just curious, like what is your sort of initial gut thought?

You know when you, when you have a person, or I should say a company, Mark Zuckerberg with meta AI, dumping hundreds of millions of dollars into buying these server farms, chip farms, attracting the best talent in the world and then just giving it away for free. Crazy, right? They're taking the scorched earth mentality, the scorched earth strategy. And by the way, I'm a hyper competitive person so I love it. Like they were behind, right? They weren't, they weren't anywhere close to OpenAI, so they're using the scorched earth strategy. I think likely what is going to happen is for a while closed source, specifically Claude chatgpt will probably be three to six months ahead of open source. But it's like the four minute mile. Like once you've seen somebody else do it, you know, it's possible. And so once you see for example like GPT4O where they have the multimodal model all like voice, it sounds super real all of a sudden your intonations and your voice can be input as an input to the model. Other open source builders look at it and they say, oh yeah, okay, well if we didn't already know about that, like, yeah, now let's go do that. And so I think there's going to be this gap, but it's going to shrink over time. And here's my other thought about who wins in the long run. If you're an OpenAI, you have ChatGPT and you're building and the Models themselves are becoming commoditized quickly. And so you have ChatGPT and the value is with all of the developer tools that you build around ChatGPT. However, all of those developers tools that you build around it are only applicable to ChatGPT. So if one, a developer, a business wanted to use a different model, they couldn't. They are completely locked into the OpenAI platform. Whereas with Meta or if somebody builds a suite of developer tools on top of open source models, you could swap out the open source models as much as you like. You can find that perfect fine tuned model for you that is efficient, high quality, low cost. And I think that's a really powerful strategy because you as a buyer of inference, as a buyer of the model outputs. I'm not locked into a platform and so that's why I like if I were to choose as a business I would probably do the initial experimentation of whatever I'm building using one of the closed source models and then as soon as I found something that worked and my code is pretty sound, I would try to convert it over to open source as quickly as possible because platform lock in is real.

Yeah, I mean I kind of agree with you and I hope you're right. Like I want open source to win, but I'm pretty skeptical. I mean a lot of the stuff you're saying kind of assumes that GPT5 is going to be slightly better and that it, and, and also that OpenAI is not going to get to some form of self improvement before open source. Right. If they do, that changes everything. Right. And then, and the rumors from people I know who know Sam, you know, a lot of things sound like GPT5 is very, very good and that people are going to be shocked quite soon. And so I, I think that everyone's comparing, you know, GPT4 and I think that's like a really old, I think OpenAI is probably starting GPT6 right now and GPT5 is, is basically done and it's going to be way better than anything that currently exists.

So I think there's some validity to what you're saying. There was an interview that Yann Lecun, the head believe is chief AI scientist at Meta AI, gave on the Lex Friedman podcast and the way that he talked about it was kind of this path to AGI is it's not like an on off switch, it's not like suddenly going to happen. There's not these huge step functions of improvement. It's more gradual, it's more subtle than that.

But he's Always been trailing so far. So. So just because he works at Meta, he has not innovated much at all. Like he's always been behind and catching up. He hasn't actually been the one who's created new things.

You know what, you're sounding like Elon Musk right now.

But it's true, it's true. That's, it's that that is the first principles way of looking at it like he actually has.

Unless it's published.

Yeah. Which is crazy.

So yeah, that I agree. That's crazy. So it would be hard for me to believe that OpenAI has some mind blowing technology innovation that we just could not even fathom to this point and they would release it all at once. It's hard for me to imagine that scenario I like the most ahead I would guess they are is 15% maybe and then that gets, that gap gets closed within six months. So I'm less bullish on the open air long term play. Especially because models are becoming commoditized and as they become commoditized they are going to find it more and more difficult to attract the best talent, to get more funding, to get the subscribers. Right. Because you can go anywhere. And so it's just a race to the bottom on pricing and so that, that will actually help defeat the moat that they have in terms of kind of just model quality.

I hope you're right. My gut is that you're very wrong.

I feel like Sam Altman too has been like really sort of trying to set these expectations because whenever you hear him talk, he's constantly talking about like the world doesn't like massive changes. They want this stuff to kind of happen gradually. And he keeps on saying that kind of wording every time he's interviewed. So you know, I kind of lean more on the side of like, I don't know if GPT5 is going to be as big of a leap as everybody, you know, might think it is because of Sam's subtle little hints. Like to me it sounds like he's trying to manage that expectation by saying things like the world wants incremental steps. They don't want this quantum leap all at once.

I think, I think it's because D5 is going to be amazing and he's trying to calm people down.

Is he saying that because they have some incredible, incredible thing and they want to drip it out over time and they don't want to reveal their secrets or are they saying that because they don't and they want to manage expectations?

It could be either yeah, it could go either way. You're totally right. You know, I have this sort of theory that OpenAI and you guys could definitely debate me and try to find some holes in my logic here, but I have this feeling that OpenAI is actually sort of not a very good place right now. Because if you look at OpenAI, right, they've got two main things. They've got their API that other companies can go and build AI related platforms on top of, and then they've got their consumer facing product with this, which is ChatGPT. Well, their consumer facing product in ChatGPT is that's becoming more and more commoditized, right? Like it's just built into Instagram and WhatsApp and Google search. I mean, not Google very great in Google search yet, but it's built into Google Search and it's just like very, very commoditized. Anybody who wants to talk to an AI, there's a hundred different places they can do it now. So the need to go and pay 20 bucks a month to do it at Chat GPT, the value of that's getting smaller and smaller. Then you look at the API side, right? Chat GPT or GPT 3.5 or GPT 3, that was the only game in town for a long, long, long time if you wanted to use an AI API, right? But now we've got Claude from Anthropic, we've got Gemini, we've got Llama, we've got Mistral, we've got all of these other options for APIs, and a lot of them cheaper than what OpenAI offers. So their two main business models have both become kind of commoditized, right? And then, you know, you put all of the safety stuff and all of the weird drama and lawsuits and stuff, you stack that on top. To me it paints a picture that OpenAI is probably gotta do something. They either gotta have something really big with GPT5 or you know, Siri uses GPT now and when, when the new iPhone launches, it's got, it's got GPT as the Siri model and that can reinvigorate OpenAI. But right now I kind of feel like they're in trouble. But I'm curious your thoughts. Like, are there holes in that logic?

So, okay, a lot to unpack. I think for the most part you're, you're 100% right. I think from a consumer perspective, OpenAI actually has pretty big dominance, right? They, they have a pretty big moat. Just because chatgpt is the verb now.

It is AI for most people, right.

It is AI, nobody even knows of. Anthropic. But your point of view, you know, Google Search is going to have it, Facebook is going to have it, Instagram is going to have it, but it's not going to be chatgpt, it's going to be Llama, it's going to be Gemini, and it's going to be more native into the existing interaction of whatever that is. And so I remember Sam Altman, do you guys remember when GPT apps kind of came out for a little bit where you can like call like Priceline and.

Oh, the plugins.

Yeah, plugins. Right, that's what they were called. So just a couple months after launch, he said something super interesting that you, you just reminded me of. He said, we kind of realized people don't want to go to. They, they don't want to have their apps in ChatGPT. They want to have ChatGPT in their apps. And, and like that stuck with me. And that's, it's very akin to what you're just talking about, Matt, where like if Google search has AI, Facebook has AI, Telegram, Instagram, all the grams, they have AI. We're like, you're not going directly to ChatGPT. So then that leads us to the Apple partnership, potentially, right? Yeah. First of all, that blows my mind. What is Apple doing? Trillions of dollars and like, they couldn't do this, you know?

Yeah.

Like a handful of engineers in a basement can pump out a decent model. So it's like, very disappointing. As somebody who's been a long time Apple fanboy.

Yeah, but that would like, go back to what I was saying is I think OpenAI is very ahead. And when you actually see that behind closed doors, yeah, they could throw together something like GPT 3.55 is so far ahead that yes, Apple would just say we're bowing, we're partnering with you, we're going to make the best hardware, we're going to continue, but you have won the game. And so we're going to make an alliance with you. And so I think that's what's happening with Apple.

In my opinion, it wouldn't be actually the first time that Apple did something like this. Right. They, they bowed out of the search game. They. And, and Google pays Apple billions of dollars a year to be their search engine. So like, would, would it be like a similar setup?

I, you know, I don't, I don't know that that is, that is interesting.

Can you see OpenAI paying Apple?

I don't, I, I Think actually Apple's going to pay them, if I had to guess.

Yeah, it's interesting because the Apple real estate is important, obviously.

I assume with the Google thing, Apple had more leverage because they could come in and say, hey, we actually do have the talent to build something like Google. Maybe it's not gonna be as good. It's gonna be 95%. And so with that argument, then they could get it where? Okay, yeah, okay, Google, you know, Google, you pay us. But if OpenAI is very far ahead and then Apple doesn't have that leverage to come say that to them, like, hey, yeah, we can just catch up overnight. If they're, if they're not able to say that, then they have no leverage. And then I think that would result in OpenAI getting paid, I believe.

Well, I think, I think this is just a stopover for Apple. I think, you know, you look at what Apple did with intel, right? They put intel chips in all of their computers for the longest time, but as soon as their M series of chips came along, all right, bye bye, Intel. I think it's the same kind of thing. I think chat GPT is their stopover to whatever they're building.

I think that's a, I think that's a good analogy.

Yeah. They'd be crazy not to be trying to build AI. I think it'd be crazy long term to not be trying.

Yeah, yeah, for sure. I mean, look, when I First, first saw GPT4O and the voice interactions, I was like, okay, well that's Siri, right? That, that was, that is the promise of what Siri should have been a long time ago, right? And they couldn't accomplish it. And GPT4O, which is not, as you said, Nathan, it's not even GPT5. They were able to accomplish something pretty impressive, I guess. By the way, there were also rumors that it was Google's Gemini that was going to be powering Siri. So I don't know. No wonder OpenAI is hiring for internal.

Aspire. Did you see that job posting? They were hiring for like an internal risk assessor or basically like prevent the leaks. That's really the job description. Prevent the leaks. Then again, you know, Microsoft is open AI, right? And Apple and Microsoft have a long, contentious history.

Well, I would not say they are OpenAI. I mean, because, you know, the whole, like that OpenAI can get out of it when they have AGI. And that's, you know, you can, you could possibly argue they already have AGI, depending on what level and you know, depending on the definition. So I wouldn't, I don't, I don't think Microsoft has a complete hold over OpenAI. They have a lot strong influence, but I don't think. A complete hold.

Yeah. So I was just looking at it. I was just watching an interview of Elon Musk. I think it was something called Viva Tech, something like that. He called it Microsoft's open AI. He like said it multiple times. I could tell.

Like, oh, he's poking for sure.

I mean that's the same thing with like. That's the same thing with like you and. Right. Like you're just following orders. Boy. Anything like, it's like.

I, I thought.

It was so funny. Definitely gave me a smile when I heard him say that. But I, I'm, I'm on a different. I, I have a different position to you, Nathan. I think Satya Nadella is playing 4D chess and everything. Everybody Elon Musk meta AI, they're all his pawns.

Y.

Because he invests in open AI. He takes all their tech, builds it internally, builds it into every level of Windows, but doesn't call it OpenAI. Also partners with Metta on the open source. I think he also did an investment. I might be wrong on this with Anthropic. Like he basically put his chips on, on every potential option.

He basically bought inflection AI too. Like inflection AI just got consumed by Microsoft.

He's. I think Satya Nadella is just, he's blowing my mind right now. Just 4D CEO chess.

Yeah, that's. It's kind of like the classic Microsoft playbook, isn't it? What was it? It's like extend, exterminate. What's the, what's the other part of that? There's something I'm missing.

Oh, I remember that.

But that was like always Microsoft's playbook. Right. Was like to get things and then like make them a bit better and then like you basically end up like killing off the competition.

Ms. Mobile. They were strong. They were a little bit late, but they got very strong with cloud and now they're early and strong on AI. So very bullish on Microsoft. Not investment advice.

Yeah. I mean he's doing way better than the CEO of Google, that's for sure.

Yeah, yeah, that's right.

Yeah, yeah.

I mean if you do any searches right now about Google AI, all you're going to find is their blunder with their like image generation model that couldn't get the races correctly and how Google is teaching people that they should be eating rocks like Google Is, yeah, they're in a tough spot right now, I'd say.

Yeah, that makes me like. But that goes back to like the thing of adding AI into everything that you were talking about earlier, Matthew. Like, I'm not so sure about that. Like, yeah, you add AI to Instagram and all these things, I think we're gonna be seeing more new experiences versus just tacking AI on to old things personally. Because like, you look at Google, they've, they've basically just tacked on AI because they're, they're in panic mode and they're in a panic mode for a few reasons I think people are not really thinking about like, yeah, sure, like AI is going to eat their lunch or whatever. But also there's a flood of AI content on the Internet now and they're having to deal with that. And they don't have a clear answer to how to deal with that. And the, the, the algorithm leak that came out a few days ago shows that like, they don't currently have any way to deal with any of this. They're kind of like falling back on. Like, okay, who has the highest authority? Well, now it's a, it's a Reddit side. Well, okay, now people are just shit posting on Reddit. So now it's. So how do you answer that? You know, if the highest authority sites have people shit posting and stuff. And so I'm not so sure that, you know, just tacking on AI to things is going to be the plague. I don't know where Google even goes from there.

I agree with you, Nathan. I think that'll level the playing field. It'll give people a lot more options to be maybe first exposed to AI. But I agree, I don't know what it looks like, but new experiences with AI seem like the inevitable future. GPT4O keeps coming to mind. Like that experience of being able to actually just have a real conversation with AI that can understand my tone, my emotion also reflect back its own tone. Emotion that's really powerful. The voice interface, which really hasn't been tackled, seems like the most obvious user interface. Yeah.

And that's where I do hope the open source like stays up, you know, stays close to open AI. So like, like startups can actually be building those new experiences. It's not just like, oh, it's all open air.

Yeah, agreed.

Another thing that, that I followed you a lot for is your sort of coverage of AI agents. Right. You've talked about agents a lot in videos and it feels like most of the stuff that I've played around with isn't quite there yet. I mean, what are your thoughts on AI agents? Have you played with anything that is like exciting you in that world? Is there anything that's like bubbling up that you're like, all right, if you want to see an AI agent at work, go mess with this?

Yeah. So I, I'm, I'm very, very bullish on AI agents. There's, there's really like two main products. Autogen by Microsoft, which is more of a research project, and then Crewai disclosure. I'm an investor in Crewai, so very bullish on agents. And there's, there's a few reasons and I'll also answer your question. I'm bullish on agents because when you just do a single prompt to AI, you're just not going to get the type of results as if you first of all had a more complex prompt, but also allowed it to iterate with itself, to reflect on its own output, to work with other different models in coordination to give tools to them. So at the end of the day, agents to me are really two things. It's the ability to put together multiple large language models to work together which have been proven through different research papers like Reflection and Tree of Thoughts to output better results. And then it's also kind of all of the infrastructure around the large language model and the workflows that you need to bring it to a production level environment. Like I mentioned, tools, benchmarking, logging, all of this stuff is kind of all in these agent frameworks now. And so if you're building production level AI, it kind of goes hand in hand to use one of these. That's how I see it. But I also agree, like it's still very early days, the agents often don't behave exactly like you need them to. And that's the non deterministic factor at work there. I think when you're talking about use cases that work really well, automating things that are very well defined. So in the work environment, research, analysis, you know, crafting content, all of these things are use cases that agents do really, really well. Beyond that, we're still trying to figure it out. And I think the improvement is going to come because of two things. Model improvements and then framework improvements. And then as you combine those two things and they both get better, they'll kind of build off of each other and get exponentially better over time and we'll be able to automate more and more real world tasks with agents.

Do you think we'll ever actually see a large Action model.

Oh, man.

That'S. Yeah. Okay, so I. Look, I was a fan of the Rabbit device. Like, I got it. Like, that's what you're really asking about, right?

No, yeah, I want to. I want to get into the Rabbit thing.

Yeah.

So, okay, so large Action model, there's actually a few examples of it in the field right now. There's. There's two projects. There was one research paper which allowed the large language model to control a Windows, Mac Linux environment through kind of a special version of those environments. And it worked really well. Open Interpreter, that was the other one. Right. Open Interpreter was a. And now they have the one. I think it's just called One, which is like a little device, but it essentially allows you to control your computer. And that's really what a Large Action model is. It's. Can the large language model write a script to execute things on a computer dynamically? I think we're gonna have that as like a middle ground. But like, that is. That is like the stop gap to a place where large language models can just execute code directly. Like you are just speaking your command. They interpret the command and then they just write code for the end device, whatever that is. So let's say you have a smart fridge. You say, tell me what's in my fridge. It writes a script to go execute on that fridge. Now, I'm sure a lot of people who are wary of security are shuddering right now, but that I do see as the future. And I made a video about, you know, developers probably won't be needed in 10 years.

Yeah, yeah. We actually recorded a podcast episode about that concept as well, but we never released it.

I would be interested in watching that. So, yeah, I guess like right now, large action models don't work very well, especially the ones that overlay a grid on top of an operating system, because it's just hard for the large language model to predict an X, Y coordinate on top of an image. But I've seen some decent examples of it, and especially as. And this is why, again, I'm so bullish on Microsoft as they expose more of their operating system to the AI directly. And the AI can control the operating system directly, and it's a well defined interface between them. The kind of idea of AI controlling your computer becomes more real.

Yeah, yeah. When it comes to the Rabbit, I loved the concept of it. Right. Like, I liked that large action model concept of like, hey, go do this thing. You train it once at your computer and then forever. Beyond that, you can press a button and get it to do that same thing again. Right. But obviously, as we've kind of seen so far, the rabbit sort of underdelivered on some of its promises, but, yeah, we could just kind of leave it there.

Yeah, yeah, yeah. I think this is a degree of over promise, under deliver. A very strong degree of that. I. You know, you can watch Coffeezilla's video about whether it's a scam or not. I don't believe so, but certainly underdelivered.

Yeah.

Yeah. I honestly don't think they built it with the intention to scam people, but I do think maybe they got in over their head or something like that. So.

Yeah, well said.

So the whole, you know, our mutual friend Bilavel Sidhu, he runs the Ted AI podcast, he just had Helen Toner on, who was one of the board members of OpenAI.

What a get.

One of the, One of the big things that she said in that interview was she broke down a lot of the things that Sam Altman lied about. Right. You know, one of the things she mentioned was they found out about ChatGPT on Twitter.

Yeah. I think, I think the most damning One was that OpenAI's board didn't know that Sam Altman owned the OpenAI fund.

Yes.

And. And in parallel, went in front of the Senate and said, I have no financial incentive in OpenAI. Now, I guess, like, if you want to break it down to technicalities, maybe that's true, but even, even that, that's just plain old false to me. Right?

Yeah, it's crazy. I mean, to me, at first I thought, like, this is crazy, they learned about it from Twitter. But then like, the more comments I got and the more I thought about it, the more I'm like, that's actually not that really. That's not a big deal. Right. Because like, at the time the API was being used in a lot of tools. It was out there in the wild. And so basically, OpenAI made their own sort of, their own sort of tools, using their own API to show off what it was capable of. It was a research preview when they first put it out. Right. So like, if you're thinking you're just putting out a research preview, are you really running that by the nonprofit board that's over here? And on a day to day basis, probably not. Right. It's like the more time I've had to think about that, the more I'm kind of like, yeah, that's probably not that big of a deal. But the thing that you're talking about where he's Flat out saying, I have no financial interest in this. But then he does have financial interest in the OpenAI startup program. Like that to me feels very shady.

The whole thing is weird though. He doesn't own equity in the company is also. That's the shocking thing to me though. Like, why is the board not fixing that? Like that's, that's the thing they should be fixing. Like, how can you have a CEO who doesn't have upside? And like, I respect Bill Wall, obviously, he's our friend. But like Helen Toner, like, what's her background? She's basically like a political writer who has done some puff pieces for China. Right. So like, what is, like what does she bring to the board? Like, how did she get on the board in the first place is my big question. And so I, I'm not saying we should say that she's lying, but I also don't like Sam Altman saying the opposite of what she's saying. So, you know, as, as a fellow builder and also I have friends who know Sam and say he's a great person who's a very honest person. I wouldn't say that we just should assume that Sam's lying and Helen's telling the truth. Like she could be lying.

I can't remember exactly, but I think she has some background in AI ethics and, and, and things of that nature. But let' side for a second.

Yeah, right, right.

Certainly she was outed. Right. She, she was put kicked to the curb out of open AI, kicked off the board after essentially trying a mutiny. So yeah, so she's going to be biased. It's easily provable whether or not he owned or owns the OpenAI fund. So that, that's like one issue. The other issue is whether he released ChatGPT without telling them. I think madam, in, in your corner on this, like I, I didn't really see that as a big deal. He even said early on he was like, I didn't think it was going to be as big as it was. Like, I thought we were going to put out this little experiment, get some feedback and then roll out a bigger project. And so I don't think he put it out intending to be like, haha, going behind the board's back on this. I think he just didn't think it was a big deal and he just put it out. There's also the report of like emotional abuse. And I like that.

What does that mean? I mean, honestly, today, what does that mean?

Give me, give me the receipts. I need to see some documentation or Something. Because what some person might consider emotional abuse might just be a strong disagreement on something, or you didn't get what you wanted or, you know, like, I, it's hard for. I'm not saying it's not true, but.

It could be true.

Like, how could you say it is true? The only one that is just like, cold, hard fact is he, yeah, he did own open AI's fund.

I'm trying to think, like, it depends on, like, what, he owns the fund. What does that mean? Like, does he actually, like, is he someone who's signing for the fund? Like, like there's like, nuance there that I think should be unwrapped a little bit. Like, like, he could just be a person who signs for it or does he actually owns the entire. There's like different levels of this. And I mean, I think it's crazy, but like I said, like, I think he should own the fund or own part of it or something. Like, there should be some financial upside for him if he said that to the Senate. Yeah, that's a major screw up. And I think a lot of the, like, my critiques right now are about him. Like, his judgment would be the thing. Like, right, like, like, why did he have Helen on the board? Things like this. Why does he not have any ownership in the company? That's not. I don't think that's the smartest way to go about things.

Well, I think, you know, he, he's been very, very rich for very, very long. And so I think maybe when they started OpenAI, first of all, it was just like a research lab. So I don't think they ever planned to necessarily comm it. He was already ultra rich. Maybe he didn't, he just didn't have that long term view that, oh, I, I need to be financially incentivized or aligned with this company.

The mind that.

Like, long term, it doesn't matter. Like, if you reach AGI, money doesn't matter. Like, like, that's why he did, like, you know, universal, basic, basic income. He did those experiments which leads you to believe that. Yeah, he believes that once you reach AGI, money's no longer a real thing. So does it matter?

And, and by the way, Nathan, I've heard the same thing from both friends who I have who know Sam Altman, and then also just kind of what everyone's saying on Twitter, their stories, they all say, like, yeah, great guy was there for me, trustworthy, like, so, like, the, the only thing that I really have to point at is the AI or the OpenAI fund thing. And, you know, I don't know exactly what happened there. Also, I just pulled up an article from Axios from just a little while ago, about a month ago, that says he's no longer the owner or controls that fund in association with the company. So obviously, if that is true, he and OpenAI more generally, post the board export leaving, realized that that was not a good idea.

So I'm wondering, when he did the Senate hearing, was he still an owner at that time?

Yeah, yeah, he was for sure. I mean, and. And like, we can go back and parse the exact words he used, and I'm sure people will, like, if I'm the Senate, I'm thinking like, hey, we need to give him a call and bring him back here and answer some questions. Because, yeah, like, it was almost a joke. I forget who asked, but he's like, well, you don't have any incentive in the company. How is that possible? And he's like, well, I forget his exact words, but it's like, well, I have money and no, I don't. So what he was trying to convey was, well, of course I don't have any incentive, monetary incentive in the company. So thus I can make the best decisions for the company because I don't have some financial incentive, which is actually kind of the opposite of the way the world works and actually not the truth.

So I'm curious, has, like, anything that's happened recently? Right? Like, there's been a lot of reports, right. There's the Scarlett Johansson thing, which I think is probably more sort of. I don't want to say coincidental, but, I mean, there's nothing wrong with going, I like the sound of Scarlett Johansson's voice. Let me go hire someone that's got a kind of similar voice. There's nothing illegal about that. There's nothing. I even feel immoral about that. Right. Like, do you think any of this, like, crazy news that's been coming out, Sam putting himself on the new safety board, things like that, has any of it changed your opinion or perspective on Sam or OpenAI?

I. I think so. I. I watched the all in podcast a week ago. David Sacks had a lot of good points on this, and, and he said, like, you know, one coincidence is okay, but once you have these coincidences stacking up sequentially, all of a sudden, maybe they're not coincidences and Maybe, you know, OpenAI isn't as well run as. As we all thought. I think the, like, new OpenAI security committee is the, like, the most, like, just, like Surface level, like just a PR stunt completely. It's essentially Sam Altman, who runs the board, has a sub group of the board doing, having this security committee, when just last week, Ilya And John, the two top security guys at OpenAI, left. It's like, okay, so you're basically creating your own board to oversee your own committee to oversee your own board and your own company. Okay, that. I'm sure that's not biased whatsoever.

And don't forget too, the government is creating their own sort of AI safety committee. And guess who's on that board? Sam Altman. Satya Nadella. Sundar Pichai.

Right.

Yeah.

Regulatory capture is a real thing. Look, I like OpenAI. I like that they brought all of this incredible innovation and really opened the world's eyes to what's possible with artificial intelligence. But, like, yeah, I, I'm. Let's just say I'm a big believer in open source.

Yeah, I'm a big believer in open source. But I, I think a lot of the stuff you're seeing in OpenAI is what you would expect to see from a company that is approaching the most important thing that humanity could ever build. Right. AGI, like, you would expect that there would be major. The emotions would be very high. You would expect that major mistakes would be made. You were still human. You'd expect that there would be like, culture clashes of people having very different opinions of what you do with this new fire that we're inventing. Right. And so I kind of. I personally think that's what's going on. Like, he's under immense pressure. He's definitely made some judgment mistakes. Like even tweeting out her that, like, the whole thing about the voice and all that, most of that was bullshit. But him tweeting out her was a mistake. Right? That's like, that was a mistake. And so I think that's. That's what we're seeing here. And even all the drama, people leaving stuff, I see in Helen's statement, I see that as culture clash. I see that as the people who were more concerned about what AGI could do, they are now out. Right. And the people who are more on, you know, we want to build AGI, we think that, you know, net, it's a good for humanity. They're the ones now in charge of the company. And I personally think that's a good thing. Personally?

Yeah. I mean, you know, it's. It's hard to speculate what's going on inside. I think, Nathan, you could be very right about that and actually likely Are. Right. It's just like tumultuous time in the company. Like hyper growth, emotions are something.

Gotta be.

Yeah, yeah. And Matt, by the way, you mentioned, you know, using Scarlett Johansson's voice alike, right? A lookalike of the voice. I'll reference David Sacks again in the last all in podcast. He said, he's like, actually, this happens all the time. If you're a director and you can't afford hiring Scarlett Johansson, you essentially say, get me a Scarlett Johansson type. And it is somebody who kind of fits that role of Scarlett Johansson, you know, female, blonde, kind of, you know, has the same history of movies and the same view. Like it happens all the time. So I don't think they did anything wrong with that. But I also like the fact that they tweeted her and yeah, that's the.

Misstep right there, right? Is they tweeted it. He tweeted her and then the voice sounded similar to Scarlet. People put those together and went, oh, he was, you know, trying to clone Scarlet's voice. I think that was the misstep. Not actually hiring a voice actor that sounds close to her tweeting. The word her is the misstep.

Right.

So, Matthew, I'd love to, you know, hear what you think about the news from Xai that they, you know, that elon Musk raised $6 billion on 8 million pre. A few, a few people said, I mean, that the company's worth 18 billion. No, that means it's worth 24 billion. That's all. That's how pre and post money works. So at a 24 billion valuation this early stage, it's like, I think that's the largest fundraise ever, like at a, at an early stage. What do you think he's going to do with that money? Do you think this is just all about Grok? Is there something, you know, bigger that in the works or what do you think?

First of all, him raising that much money is all about his name, Rightfully so.

Right.

Yeah.

It's not like this is some random dude coming out of the woodwork building an AI company. This is Elon Musk, who has proven he is one of, if not the best entrepreneurs of all time. Okay. So that aside, I think he already tipped his, his, his hat or his cards. He said he's going to be buying a ton of CPU or GPUs. Right. He's going to be investing heavily into Nvidia cards, building out. I hopefully I'm saying this right. He said the biggest GPU cluster ever.

You said 100,000 H1 hundreds, I believe.

Wow. So I think that is the right play. So a few things that is the way to attract talent. Right. Because the GPU is really the bottleneck. So if you have the gpu, you can attract the best talent who could hopefully build the best models. I'm hoping, and I am all for more competition in the space. Closed source, open source, I don't care. Competition is good for the end user, the consumer. So I, I'm very, very bullish on using all of that money for, for buying the compute. I really like that strategy. I. They. They have a data set that is incredible that really nobody else has. Right. OpenAI has been announcing all of these partnerships. So they're trying to get all this data. But like XAI has Twitter's data. That's, that's a crazy.

And possibly Tesla, then neuralink and then SpaceX and that's.

How do they, how do they, how does X. I relate to Tesla? And that's actually something that worries me as a Tesla investor because.

Yeah, me too.

Elon is kind of holding the company hostage right now. If they don't approve of his comp package, which is essentially he wants to own 25 of the company, he's going to go do AI somewhere else, which.

That is rightfully so.

Okay, so, so him holding the company hostage is not right. Like.

Well, yeah, I would say that's not right, but his comp. The fact that they're trying to hold back his comp package. So he basically did a deal where if you look at the video clips from, I don't know how long ago was it when the comp package, was it like, was it already like 10 years or something? Was it five years? It's. It's been a long time. But anyways, when, when the, when the comp package was introduced, like, there's video clips of saying that the goals in the comp package were so ridiculous that like, what the hell is he doing? He's crazy. There's no way that anyone would ever reach these goals.

Right.

Like, if he doesn't reach it, he's not going to get paid. He's insane. And, and so of course if you reach those goals, he should be compensated for reaching those goals. It's insane that over a decade now that he's created this company into this huge behemoth in the industry and then now some investors are trying to say, oh, we don't like your politics or whatever, so you, you shouldn't get paid for what you did the last 10 years.

Well, they did what?

The crazy.

The comp. Right.

That's crazy.

It's absurd. It's absurd. He is. Yeah. I think you, you, you told that story perfectly. It's easy with 2020 vision to look back and say, like, wow, you made so much money. An absurd, crazy amount of money, which he did. But his initial compact, like Tesla was nothing when he joined. Yeah, it was like a, it was, it was like an off the shelf electric motor put in. Oh, what, what are those little things? The Lotus, right? Yeah. That's all it was. Like, he's the one who turned it into one of the most valuable companies in the world. He is the one who changed the car industry forever. So holding the company hostage hurts me as a shareholder, though, so I don't like that. But I understand and I want him, like, go ahead. If you want 25%, fine. Like, if you want to make every decision, fine. But then if he goes and he takes AI and he goes and builds it somewhere else, X AI without kind of integrating it into Tesla, then all of a sudden, what, what's, where's Autopilot going to be? And like, and then, yeah, Tesla's just a car company. All of a sudden their valuation is going to, is going to plummet because it's not this vision anymore, it's just a car company. And there's a lot of car companies out there. So I, I don't know. It's, it's.

Yeah.

What do you think?

Yeah. You know, there's a reason that they say, like, Silicon Valley pirates, right? Like my friends Ollie Murty, who had this company called Peanut Labs back the day they used to do this big pirate cruise every year and tons of Silicon Valley elites would come out to it. We'd all dress up like pirates, right. Every year there, there's a reason that that kind of tradition exists, right? Is it like people who start big companies, they often break rules, right? And they often do things that from the outside, people be kind of shocked about the stuff that actually goes on that, you know, and there is kind of pirate behavior of like going off to get the treasure and doing whatever the hell it takes to get it and breaking rules along the way. And so I, I think from that perspective, like, yeah, he's being screwed over at Tesla. So I'm not shocked at all that. Like, yeah, there's some rules that he can't hold them hostage, but he's trying to do that. That's not shocking at all.

I mean, you can, you can look at other Founder led companies versus non founder led companies that maybe previously were. Let's look at a few, a few examples. Meta AI very much founder led. Zuckerberg is in charge. He still has those super, super shares as far as I remember. And so he's able to invest a ton of money into VR, AR and then on a dime, when the market doesn't like it, kind of turn the ship around, do a 180 and now they are like booming with AI and so he's able to do that. Now you look at Google where Sergey and Larry are no longer kind of as day to day as they once were. And Google, you can see they like, they were slow to get AI. They literally made the paper transport or attention is all you need, which was the defining paper that all of the current wave of AI is built on. They couldn't productize it, couldn't commercialize it. And even when they did, they stumbled, they had the woke AI and it's just like, like it's so slow. I think they're finally starting to get their act together, which is good. But like it's very clear when you have a founder led company and somebody who can make decisions quickly, somebody who could, you know, break rules, not, you know, not break laws, but break the rules, kind of break, break the mold as a, as a Tesla shareholder, give Elon his money, like just let him make the decisions and if he's wrong, hopefully that he, he doesn't get paid as well like that. I think that should just like he, he's proven he could do it, so let him try it again.

I mean that's, I, I think we covered so much ground on this episode and talked about so many things. But yeah, this has been awesome. We should definitely do a round two at some point if you're open to it.

I would love to.

You know, thank you so much for, for hanging out with us. Before we wrap though, like, where should people go? I know you've got your YouTube channel, you're doing stuff over on X. Where's the best place to go check out what you're up to?

Yeah, definitely check out my YouTube channel. Matthew Berman. Just search it in the search bar and then matthewberman.com if you want to check out my newsletter. Awesome.

Well, thanks again for, for coming on and just sort of nerding out about AI with us today.

Yes, sir, Anytime. Guys, seriously, this is fun.

Sam.