EP7: AI Photoshop War, Claude.AI Controls PC, Apple’s AI Push, A16z GPU Move, & Humane's Price Cut

In this episode of "AI with Friends," hosts Marlon and Adrian dive into the evolving landscape of AI technology and its implications. They explore the integration of Pinecone databases and vector databases, discussing their significance in the AI-centric world. The conversation shifts to the impact of AI on creativity, with a focus on image manipulation and the potential threats to artists. They also touch on the advancements in AI models, such as Anthropic's new capabilities and Apple's upcoming software updates, questioning the balance between innovation and privacy. The episode further examines the challenges faced by startups like Humane, highlighting their recent product struggles. Finally, they discuss Andreessen Horowitz's strategic move to support AI startups with access to Nvidia GPUs, emphasizing the changing dynamics in venture capital. Join Marlon and Adrian as they navigate these complex topics, offering insights and perspectives on the future of AI.
----------------------------------------------------
Welcome to AI With Friends, your weekly launchpad into the world of Artificial Intelligence. Hosted by Marlon Avery, a pioneer in GenAI innovation, alongside Adrian Green, VP of Engineering at LiveNation, and Sekou Doumbouya, Senior Staff Cloud Systems Engineer at Pinterest, this show is your go-to source for all things AI.

Our hosts bring diverse expertise—from AI strategy and tech innovation to industry leadership. Every week, they break down the latest AI trends, interview top experts, and simplify complex concepts for AI enthusiasts, entrepreneurs, and tech professionals alike.

Marlon, Adrian, and Sekou combine their unique perspectives, whether it’s Marlon’s collaborations with tech giants, Adrian’s leadership in global entertainment engineering, or Sekou’s cloud systems expertise. Together, they make AI insights accessible, actionable, and exciting.

Tune in live on Twitch & YouTube every Wednesday at 9:00 PM ET, or catch us on all major podcast platforms.

Here’s what you’ll get:

Cutting-edge insights from AI leaders
Real-world applications of AI technology
A vibrant community of forward-thinkers

If you're ready to stay ahead of AI trends or spark your next big idea, join us each week for an hour of engaging, thought-provoking content.

Subscribe now and become part of the future of AI with AI With Friends!

Chapters

• 0:00 Introduction
• 4:19 Discussion on Pinecone and vector databases
• 8:49 Concerns over AI's impact on image manipulation
• 11:28 Creativity vs. Automation in the AI era
• 15:56 Timbaland's use of AI in music production
• 17:48 White-collar vs. Creative benefits from AI
• 19:31 OpenAI's hiring strategy and legal implications
• 23:32 Anthropic's new AI model and its implications
• 28:14 Pros and cons of AI automating computer tasks
• 31:06 Cybersecurity concerns with AI advancements
• 33:39 Compliance and security in AI usage
• 37:02 Apple's upcoming software updates and AI integration
• 42:35 Discussion on Apple's strategy and innovation
• 48:57 Advanced voice integration in Apple devices
• 52:49 Humane's AI Pin Price cut and its challenges
• 59:54 Comparison of on-device LLMs and cloud models
• 1:07:27 Tesla's robot demonstration transparency
• 1:09:01 Andreessen Horowitz's Oxygen program for AI startups
• 1:29:28 Conclusion and episode wrap-up

Transcript

Marlon Avery: Hey, hey, hey, hey. What's going on, Adrian? 

Adrian Green: What's up, Marlon? Doing good over here, man. No complaints. You doing all right? 

Marlon Avery: Hey, man, I'm doing great, bro. As you know, man. Been working on some stuff and building some stuff, you know, in this AI centric universe that we live in. And so. Yeah, man, being going great, bro. How was working? Any cool projects throughout the week? 

Adrian Green: I have to say the coolest thing that I've been working on is, you know, it's been really a labor of love. It's been really. I've been working on integrating this pine cone database into what? Into what I do. So I've been playing around with different embedding models and different natural language models in order to ask it questions, and I've been experienced some success, so it's. 

Marlon Avery: Pretty exciting for the people. Real quick, maybe we didn't do this before, man. Explain what Pinecone and vector databases, you know, are and why we need them in this universe. 

Adrian Green: Now, there is a much better authority on me than me out there on this topic. But in short, what a vector database is, is it's a data store for embeddings. So it'll be chunks of data from a source. So this could be text. Text will be, like, split up into chunks across, you know, entry items in a database, and a natural language model will be able to interact with the Pine cone database to pull information from it and generate summaries using its brains. That's about as far as I know about it. But what it is, is it's a convenient data store in order to store the information that you have privately, not shared with anyone else. 

Marlon Avery: Yep. Yep. No, exactly. Exactly. Well, yeah, man. Well, uh, you guys, as you know, man, this is AI. We're friends. We've been doing this podcast now, man. This is episode seven. And so, man, we are practitioners. I stuff here in the space. We are builders. We're developers. We know, we're engineers, and we create this podcast because we've been talking around this subject for, like, the last year or so. We just been doing kind of group calls, um, and stuff. And we said, we kind of got to the place, and you say, you know what? This information is probably, uh, it's probably beneficial, and I think to the rest of the world and stuff here, and. 

Adrian Green: So we should help. Yeah, yeah. 

Marlon Avery: We should put it out and stuff here, man. And so, uh, yeah, throughout this, throughout the call and throughout the segment, we, uh, we talked about some of the things going on here in the AI universe. We talked. We gave our perspective on it. You know, we're saying we have a unique perspective because we're building, you know, we're engineering, we're developing solutions every single day. And so with our practitioner stuff here in the space, man, we kind of want to share with y'all. And so, uh, yeah, man, let's, let's kick it off and everything. So here we go. Uh, the verge, uh, put out an interesting article. Um, in quotations here, they said, we've had Photoshop for 35 years, often used to downplay, you know, concerns over AI's impact on image manipulation. He said that, he wrote that as challenging this comparison, pointing out that while Photoshop requires technical skill and significant time, today's AI tools, like those in Google Pixel reimagine feature rooms are reimage feature allows users to effortless generate sophisticated fake images, which illustrates how AI is making image manipulation easier and more convincing, raising alarming questions about the trust and what we see. Unlike traditional tools that we've seen and stuff in the past, a now enables anyone from a smartphone to create a deceiving image in seconds, eroding the credibility of the visual media as AI becomes more accessible. The question isn't about its capability, it's more about the implications and trust and authenticity in a world increasingly driven by AI visual content. Adrian, man, what do you think? Man, are we getting into this AI centric world, particularly on the multi model side of images, where it can become a tool for bad use? And so what's your thoughts on stuff here? 

Adrian Green: Well, you know, I feel for the artists, I feel that for the people who are maybe a little bit worried that their graphic design degree or, you know, their degree that they have in the arts could be threatened by, you know, AI encroaching more and more into the Photoshop area. So first and foremost, I feel for them, but you know, really I think the creativity is going to win. At the end of the day, there's basically creativity versus automation. And I think if you have a vision, you have a voice that's unique, you will still and unique and it resonates, there's still room for you to, you know, really make it out there. You know, with AI, it's using the data that's already there, it's using the images, it's, you know, in referencing images that have already been created that are already out there. So, you know, it can only do so much. So what I think is what it's going to end up doing is being an assistant more than a replacement. So there will still be required a lot of, you know, probably using, you know, I don't want to use this word too much, but fine tuning, you know, really adjusting what is generated by Photoshop to clean it up and make it more presentable, make it more authentic to what you're trying to do. So the more that you know how to use Photoshop and, you know, you have a strong technical background in that, I think that you'll be good because I compare it to coding, the state of being a coder right now. You know, the code codes itself, but the code that codes itself still needs to be checked. You know, we haven't fully crossed that threshold into, you know, let it rip. You know, make me a full application and deploy it. We're almost there. But having a strong, having a strong coding background is going to ensure that, you know, the quality of what you're making is there. 

Marlon Avery: I was watching the music producer Timbaland the other day. He started to create some AI generated songs. 

Adrian Green: Oh, yes. You know, too. 

Marlon Avery: Yeah. And, you know, he really, really, you know, started to like, you know, dive into it. And I think it was interesting to watch his experience because he has the knowledge, the skill set to understand what fundamental things you need to do to, like, create a song, if not a hit. And so giving someone like this in the creative world, this becomes a superpower for somebody like him. And I think comparing this to artists who are looking to use these tools and everything on the imma side of things, it also can become a superpower, you know, to them. Now, I do, I do understand the concern stuff. You know, I understand the concerns. I understand the concerns. And then, because, like, you know, there's still a big question mark on how companies like OpenAI Anthropic, how they train their image model. It's still rumored that it's been trained on public and private images along with companies, along with other image model companies. I think it's really interesting, too, because how this labels towards Adobe with Photoshop, because apparently Adobe is right now the only model that is safe to use. On the b two B side, they train their image model on their own images. And so it's rumored right now that they didn't train their images on public information and public data. So it can't come back to you and kind of like, you know, kick you back and stuff in the long term, you know, and stuff and everything. But I think this, I think this topic kind of goes with everything. I think, I think it, I think the, the multi model style with images I think is aligned with, you know, researchers and reporters, instead of being concerned around, you know, it affecting and replacing them, you know, I think it's the same thing for video. I think it's the same thing for music, you know, and all these things. But there are individuals, companies who are taking and using this and making it a superpower of those. And so I think in this world, as long as you're playing defense, you're gonna lose. Yeah, I think this. I think this, this is too, this is too. I mean, this, this thing is rolling. I mean, it's the, this is like the old kind of cartoon. This is like the old cartoon of. Yeah, this is like the old cartoon of the snowball kind of rolling down the hill, you know, and it's kind of getting bigger, bigger, bigger, bigger, bigger and everything. And there's no stopping it. So at this point. And so I think, you know, if you switch over to office and figure out how to use this along with your knowledge and skillset, I think you'll. 

Adrian Green: Come out on top for sure. And this is the age of the hackers, you know, it's the age of the people that get in there and try and figure it out. A lot of these models are free and open source, and you can get up there and use it. And I. It's not surprising that Timberland is one of those people, you know, being a Virginia, Virginia beach native. Yeah. Went to school with his brother. No big deal. High school in the Lambo, no big deal. But I think that's an example. You know, a lot of the, or, you know, him and Pharrell, too, these are people who really taught themselves how to do what they do. They had great mentors along the way. You know, Timberland had Devontae swing and Missy Elliott and everything in his circle. So that was, that was a valued resource there. But it was the, you know, initiative that he took early on to really, you know, figure it out himself. And it just, it does not surprise me that he is interested in and using AI to make beats, you know, be interested. I'll be interested to see if he releases anything that's been, you know, augmented by AI. 

Marlon Avery: Yeah, I mean, here he was saying on his life, he was like, 2025, he said he's, he's pushing it hard. Yeah, he's going to push out a lot of things, you know, as well. You know, I think what's interesting, too, as well, you know, a couple years ago, as these models and stuff were being developed, the notion was that it was going to tackle white collar jobs and tasks are first and then creation, I'm sorry, not creation. Creativity second, or creativity last. You know, so I say, and now we've seen that flip. You know, I think, I think, here's an interesting question. Are the white collar individuals benefiting more right now than the creator? From the creators, you know, from artificial intelligence? 

Adrian Green: I mean, I would think the white collar people who know how to use AI systems are for sure benefiting more than the creators right now. Because it's using. They're using it. Yeah. Yeah, I would say so. Because if the creators are of generating AI generated content, it's clearly AI generated content. It's something that's going to be flagged and lambasted, potentially, of being very not creative. So it's not, as far as in the creative space, it's not really there. But if you're using it in a white collar position, say, like document summarization, writing emails, all of the things that you have to do really peripheral to what your major job task is. Yeah, that's an invaluable resource right now. 

Marlon Avery: MIT and Stanford did a research, and they said that they recognized that individuals who are using tools like chat, WT, has already seen up to a 20% increase in productivity. And this is like, in your, you know, your workforce, like your labor, um, and stuff. And so, yeah, I think that. I think it's interesting, and I think, uh, yeah, it's interesting to see that the, the, it feels like, it seems like that the workforce is benefiting from these tools more so than the creators and stuff right now. And so it feels like the creators are on defense and they're trying to figure out, you know, within reason, too, with understanding too, as well. They're trying to figure out how to, how to protect their ip, you know, trying to figure out, you know, okay, did this model use my information, you know, my, my creation, you know, and stuff? And so, and, which I understand. I don't understand me. I mean, that the New York Times has been extremely public about, you know, their lawsuit suffocates opening eye. And so, yeah, I think this is interesting that it feels like that the workforce is benefiting more so than the creators, you know, from these Tools. 

Adrian Green: But now that you've mentioned the New York Times, but this isn't a topic I don't think that we're going to discover, but just we can touch on briefly, is that aren't they getting New York Times and maybe the Boston Globe and maybe even the Daily bugle from Spider man? Aren't they getting some people from OpenAI to, you know, come by and introduce them, little AI representatives. And I think they're getting broken off some money for that, too. 

Marlon Avery: So the other way around. So OpenAI has decided to hire writers, publishers, individuals who potentially work at these companies, you know, who work at these companies. So like in New York Times or people who worked in the companies previously to come in and, you know, write, you know, on their behalf or anything so they can train their models. And so how do you write, you know, like a publisher from New York Times or so and everything? So, yes, this is interesting because it's a brilliant play by opening up. This is a workaround. This is a workaround. So the is a public has been, it's been a few large companies who are suing, you know, OpenAI. And I understand, you know, they are training their models off of context that it got from their, I, you know, from their website, you know, and so, and a lot of these platforms, you know, have a paywall, you know, to their websites. And OpenAI has figure out ways or some roommate support. They figure out ways how to kind of get past these paywalls and everything. 

Adrian Green: Oh, there's totally no way to get around a paywall, right. 

Marlon Avery: So they figure out, they figure out ways how to get around these pay paywalls or anything to scrape, you know, their websites. And so New York Times, they, they figured out that, hey, you know, a lot of the information that your models were outputting comes directly from, you know, our ip. So what opening I decided to do, and they've done so along with, in aspects of like voice into as well. I'll talk about that in a second. What opening I had decided to do is they decided to, I think they allocated, they're going to allocate like $10 million, like, to this fund. And so basically they're going to allocate $10 million to hire individuals to come in and write, you know, for them. And so it's a beautiful workaround because when you take it back into the court and the court says, okay, cool, now show us, you know, the context of where this Information came through on your models and everything. Now openly I can say, sure, anything. We hired Bill, who was a previous worker or current worker at New York Times, and Bill came in and wrote, you know, for us, x amount of hours and we took the Information and we trained it, you know, thing on our model. And so now when you have that type of like flow, you can no longer sue, you know, you can, you don't have a lot to stand on and Stuff anymore. If you're New York Times, you know, if you're one of these, you know, large publishers and you don't have to stand up, because what, what creates a large publication like the New York Times is the individuals like the writers. And so opening Eye has found a Workaround and say, you know what, we'll just go hire the writers to come in. They did the same thing. This is why, this is why Tetrip T and the only on the advanced voice, I can do accents like jamaican patois because they came in and they just hire a bunch of Jamaicans to come in and talk, you know, coming and talking to the model. And it trained their models and stuff off of that. And so, you know, this is a, it's a workaround. It's a brilliant workaround. I know New York Times is super annoying, and I know they're pissed, but it's a brilliant workaround. 

Adrian Green: Yeah, it's an interesting approach. I mean, I really feel like I would be more, it would be more common. I wouldn't be as surprised if OpenAI was just offered New York Times, you know, some money to, you know, maybe either license or buy their, so they. 

Marlon Avery: Did, they, they partner with a few. They partner with a few, a few publishing companies. They did partner everything. And so basically, that's why I say this workaround is intended to not have a long term, you know, lawsuit or so forth, maybe. And so. 

Adrian Green: Yeah, exactly. Exactly. Makes sense. Makes sense. 

Marlon Avery: Yeah. All right. All right, so here we go. Moving on. All right, baby, go. So anthropic is doing some interesting things over there. Kyle Wiggers from TechCrunch, he wrote in. So the latest innovation from Anthropic, they recently launched and upgraded an AI model cloud through which now makes the AI different in its ability to interact with your desktop applications, performing tasks like clicking, typing, and even handling more complex software actions through a new computer. Use the API. This now AI model can emulate human actions on a PC, potentially automating tasks like email replies, research, or even managing apps, um, by itself. While this development is exciting, the article highlights that the potential risk, uh, there's potential risk. Early test shows that the AI struggles with some basic tasks and is even gets more advanced that the growing concern is, will be proper. It would be misuse, especially in a more sensitive environments like booking systems or using personal, uh, personal data. Um, anthropic ensures us that is taking safety precautions, but it's clear that the tool can disrupt how we interact with technology, blurring the line between human input and AI driven automation. Adrian, man, you know, anthropic. Now, this new model can, you know, can do some things with your computer, of course, with your permission, you know, and stuff. Everything. But. What's your thoughts, Mandy? There's a lot of pros and cons to this. Talk to me. 

Adrian Green: You know, it's. It's great. In short, I appreciate it. We've used. You know, we've used the agent approach in our own projects, and you know where that's gone? Where that's gotten to since the beginning of, you know, first using it with lang chain. Leaps and bounds. Leaps and bounds. 

Marlon Avery: They're taking blows right now. Lane chain. It's like they were. They were. They were super. It was in the front, you know, they was leading away, you know, as a framework, and then just. Yeah, go ahead. 

Adrian Green: It's still valuable for its integrations, like, some of the tools that I've used. I think I'm going to be, you know, probably off of link chain maybe completely in the near future, but, you know, I still see some integrations on the back end out there. It's still, you know, but, yeah, this other companies right now with these agent. You know, agents are really eating their lunch. I. As far as the automating of your actual computer, I feel like that's. That's great. That's great. I don't know exactly what I would use it for other than, you know, outside of what I use agents for now. Like, I probably wouldn't need another. Another that's on there. What I'm thinking is, the anthropic model is going to most likely integrate directly with the. You know, I use the cursor ide so I can maybe, hopefully spawn an agent through that interface, connect to an agent, and then have it work something similar to how relet agents work, where you get a working window and you converse with an agent to build the application. Talking with Sekou about this is fascinating because he's just like, you know, it's a different way of coding. You know, it's giving direction. It's. It's. But you're still building the application, so it's exciting stuff. I think that there is, you know, probably the immediate use for it. I don't know if you've ever used this, but, like, you know, those tools that make it seem like you're actually logged in at work. So, you know, little mouth shake, little, you know, the mouse. Mouse movement tool. Yeah, that, you know, with the advent of something like this that controls your PC, that type of oversight may be out of business. Whoever's doing that oversight may not, you know, be as effective. 

Marlon Avery: Yeah. So I'm gonna give you the pros and cons. Let's start with the cons of this. I don't, I don't see this at scale in the near future, even from a basic task like email. I mean, I mean, they already said that they're already starting to see concerns, you know, stuff with it. I don't see this on, on the, on the automation side in the near future. I mean, as you know, you know, there's still high error, you know, rates of just building simple, you know, AI solutions, you know, and stuff. And so the, the understanding that it will start to, on a full automated side, we're talking now, we're talking about understanding size limits when it comes to text based solutions. We're also still summarizing you're implementing agency with that. We're also still, we're talking about compute along with it as well. We are still trying to figure out a, you know, the agent infrastructure of, you know, implementing, you know, solutions like this. So I think it is, this is a more, so in my opinion, this is like, hey, this is what's to come. You know, this is not a product that's going to release next month. I'll be very surprised if they release this like the next six months in beta. Sure. You know, they'll, they'll probably, they'll probably release it to like a few people or anything, but I'll be very surprised they release this in the next six months just because there's so much to figure out and stuff in that, in the area. 

Adrian Green: For sure. It's nothing to keep an eye on for sure. Because when it comes to interacting with a Windows computer, I feel like I, you know, I definitely want to keep updated on the news as far as the security considerations here because, you know, how easily corruptible or, you know, you can really have something running on your computer that you know, is malicious but somehow attached to or activated by the AI on unbeknownst to the AI, you know, particularly like say if it's running some executable in order to do some tasks, but that executable has been hijacked in some way. 

Marlon Avery: I tell you, this cybersecurity is going to skyrocket as an industry man. It is going to be so many different new solutions and implementation and things to figure out and how to solve those things and how to protect those things because the ability to incorporate a DDoS attack inside of someone as agent. Yeah. And then one of the things that still concerns me, like we still haven't, we still haven't talked about at scale how we're going to protect consumers data, you know, how to protect their private information. I mean, I haven't seen one company talking about, you know, the infrastructure or implementing something like blockchain, you know, and that's a technology and solution that's just sitting there, you know, and is waiting to be. And I, and I think, I think, and it goes back to what we do as a country. You know, we always, I said this before, we always build the car before we build the seatbelt. You know, man, I think until these companies start getting hit with like some consistent 500, 700 billion dollar fines, I don't think they're going to put kind of put this atop of mine. And so I think it's more, so this goes back to what we experienced consistently. It's like they're trying to, they're trying to make sure, you know, the investors, you know, you know, in the market and stuff is happy and satisfied stuff with these things. And so, yeah, I think those is my base concern for sure. 

Adrian Green: The compliance industry too is, you know, forming around this topic as well. So if there's, you know, when we talk about security, when we talk about these things and we talked about the, you know, the cybersecurity, you know, industry really being, you know, this, they're going to have a lot of work when it comes to around the AI, but what they're going to develop the best practices and how that's going to translate into the compliance side of things which, you know, is coming down the pipeline. If a company is using AI, they have to abide by certain compliance rules. We're talking about that, you know, the EU and California. Yeah, they're going to be the, hopefully the guardrails on, you know, this, on using AI locally on your, you know, especially a work computer. 

Marlon Avery: Yeah, yeah. And that actually know what, that's interesting point what you said with that too, as well. It's like personal use. I can see this happening. I, I see this is a big issue in a work computer. 

Adrian Green: Oh, for sure. 

Marlon Avery: You know, being, being able because also to one of the things that was it Coinbase, I think it's Coinbase or it's like a, it's like a, it's like a company that Coinbase invested in or something like that, but no coal base was involved. I know Coinbase is starting to work on autonomous payment agents and so the ability to assign agent, an agent a give to get them access to your banking or car. And once it makes certain, once it makes certain steps, like, so the example is you'll be able to eventually create an agent that will go surf the Internet for that sweater that you've been looking for. And once it finds it through, you know, serving through, you know, websites, platforms or even marketplaces, you know, then it will make that purchase on your behalf. And so I can, I can see us getting to that place and getting to that world. There's just, it's so many data and so many privacy and it is, so many things can go wrong, you know, stuff, you know, in between those steps. And I don't think a lot of these companies have that top of mind. They just want to build a solution and they want to innovate. But I think there's so many things that can go wrong. And unfortunately, you know, typically, you know, the consumer takes the, the heavier hit of that, you know, you know, versus the market. 

Adrian Green: Yeah, yeah. The companies too. And if they, if it's in the case of a data breach or something, you know, like on, in that vein, they're going to for sure take the hit. 

Marlon Avery: Yeah, definitely. All right, here we go. Brian Heater wrote, he covers the latest Apple software upgrade is coming up. So the upcoming features will include iOS, ipadOS and Mac OS system. Apple intelligence set is set to roll out soon. It's going to integrate tools like chat, WT for advanced questioning, questioning and answering image creation through its image playground and its new capabilities for Siri. So these features aren't just flashy add ons that a representative from Apple has said. So the strategy is to stay competitive in the AI landscape by enhancing everyday productivity from generating original, original emojis. It's funny they put productivity and emojis in the same sentence anyway, from generating original emojis to summarizing articles and creating image based on user promises, this update marks a significant shift in how users will interact with their devices. However, Heater notes that the apple is being cautious, particularly around its ethics and privacy as these new tools cause race concerns around blurred lines between human creativity and machine generation. So here, I'll start it off with this one major. 

Adrian Green: What you got? What you cooking up? 

Marlon Avery: Okay, so this puts me back in the landscape of what the, so you remember, remember when Steve Jobs, when they had safari, remember when Flash was a big thing on web browsers? 

Adrian Green: Oh, yeah. 

Marlon Avery: And Steve hated it. You know, he hated it. I mean, he wrote that article. He wrote that article, you know, basically about Adobe's flash of like why it's just not. It's not good, you know, and why. It's just never gonna work. And, you know, he. They decided to, you know, not include the latest flash, you know, update within safari. And for a time period there, you know, safari experience and interaction was kind of weird because you were just kind of like, you'd be scrolling. You have, like, this big gray, you know, section in the middle of, you know, your experiences like that. But they keep building and iterators around it, which, you know, they eventually start figuring out, I feel like this is the opposite of that. Okay. I feel like apple in this area is just building to stay with the times, and that's typically not like them. You know, I haven't seen anything in this area that is unique to apple, and I'm not even gonna say set them apart anything, but it's, like, unique to, like, the apple. Wait. I mean, this. This feels. This feels very much like the same thing that Google is doing, you know, and it's typically, Apple typically is a company where they haven't innovated in a long time. And that's typically. And I've figured it out, they haven't innovated in a long time. What applicants it typically does is they wait for the rest of the market to release products, you know, features, and they wait for the market to respond to those things. And then as you go, they usually go, do it better. Like, for example, that was like the. The fingerprint lock on your phones. And so, like, I remember Sam's. Sam's Samsung. Samsung's was, like, there. It was like locking people out of their own phone, and the only way you could get back into it was like you had to, like, you know, wipe the entire phone, which you lose your information if you haven't. So it was like a bad experience. And so. And then Apple came out with the, you know, the fingerprint, you know, lock and everything. And it worked. And it worked well and everything. And so. 

Adrian Green: You mean the face, the face recognition or the finger? 

Marlon Avery: No, fingerprint first. Indian face came next. Okay, so, yeah, it was a Samsung, you know, Android. They. They did the fingerprint first, I think was like the. It was on the galaxy note or something like that, whatever, maybe. So they did the fingerprint first, and. And I had it going, and it was like this new thing, man, this is great. And I can get my phone like that. And then it wasn't working properly, you know, so people start to just remove it from their phone, remove the feature for their phone, because it was starting to lock people out you know, all the Android phones. And so literally, like a year or two later, Android, I mean, I'm sorry. Apple announced it and, you know, it was little to no problem stuff, you know, with it. And so that's what I mean, worse. Like, typically they don't innovate. They wait to the market, responds to something, and then, you know, they incorporate it, you know, with. I don't, I don't feel like there's ever some data around how people are interacting with these tools. I mean, yes, we, we're using chat TPT every single day. The image, you know, things. Yeah, that's, you know, at least for, you know, I host workshops and I teach people, businesses, and organizations and governments how to implement these tools. And so it's just like the image thing that's every now and anything, you know, for, like, the everyday person. But the text generation side, you know, is something that we're using every single day. But they're not integrating text generation. They're just, you know, they are simply, they're, they're implementing chat tribute, you know, stuff with that. And so that's why I said, I don't feel like this is, it feels like they're just kind of, like, almost keeping up with the time stuff right now, which is typically not the Apple way. And so this is interesting, you know, this is interesting, you know, to do so and so I kind of get it why you would do so, because it keeps people, I don't know. No, that doesn't make sense either. I was gonna say key seapo on the device, but I'm using chat GPT on all my apple products, you know, and so I don't, it just, it's not clicking for me, you know, why you would do this and everything. And so somebody like Facebook, I get it because, you know, Facebook just has to stay with the times because they're always one big scam away, a woman thing away from being counseled, you know, as, you know, like, whatever it may be. And so, like, yeah, this is an interesting play on Apple's part particularly. And maybe the only, the only reason why I can see this being a play for them is for them. They get, they at least get the data on the back end of this to understand what people are using these tools for, then to go build their own. And so I, I think that's the only reason why I think they continue or, you know, announce these things. 

Adrian Green: Okay. Okay. That's a, that's definitely of, like, perspective. And, you know, I have to, have to take another really viewpoint on that, which is, which is quite the contrary. I love it. I love it specifically because this is an AI model that's going to run on your device. I have been downloading and using models locally for a couple of weeks now, specifically. So outside of chat GPT, and I have found that it is faster and more accurate to what I want given a context. So if Apple, which, you know, Apple is the New York Yankees, you know, Apple is the, you know, the Dallas cowboys, you know, they. It's easy for me to just not like them. This is what I'm saying because so many people do. But for this topic, I am a fan with what I know that, you know, having a local LLM can do with predictive text or on device. 

Marlon Avery: LLM? Yeah, on device. 

Adrian Green: On device. Yeah, yeah, on device. So, you know, when it comes to image generation, for example, you know, I would use that a lot if it was something that operated at that at a speed of, that was good. Faster than when I can get with a cloud service and I can insert them in the chats. I mean, I have, you know, so much expression that I feel like, give me an example. Proper. 

Marlon Avery: Give me an example. 

Adrian Green: I need to. I'm chatting with my sister, right? I need to generate an image of, you know, my sister is a cook, so we talk a lot about cooking stuff. So, you know, I want to generate an image of Anthony Bourdain, Rachel Ray and whoever Paula Deen in a cage match or something like that. That would be something funny and contextual to the conversation that we would be having. We're being critical about something. Like I would. If it was something fast that I can generate locally and just share, I would love it. And with the added benefit of potentially this getting smarter as I use it. So maintaining the context of what I give it in order to make the responses in the future even better. And it's on device. Yeah, I'm a fan. I'm a fan of it. I think it's a good move and it does seem kind of trite at this point, but the advantage of the local installed on your computer model is I think it's there. I think it really is there. And for Apple to take advantage of that, I think is a good strategic move. I also like the news that they're not, they're bringing, you know, some enhancements to Siri back. So we're not throwing her, you know, that's advanced voice. 

Marlon Avery: You know, that's real. That's real time API. Yeah, it is. 

Adrian Green: But it's local, but it's local and again, it's okay. 

Marlon Avery: Okay, so let's dive in that example, advanced voice. I haven't used Siri in forever. I use Siri now for like basic functionality of setting alarm, a timer, you know, I think I stopped using it to like call people, you know, even text people. I stopped using it a long time ago just because I wasn't having a good experience and stuff with it. So you implement advanced voice, real time API into Siri. The only thing I see what's happening is now you are saving me one or two steps. So if I open, I'm opening eyes on my, on my home screen. So I open a home and then I click advanced voice and I'm using it there versus maybe now just a button, you know, stuff on there. It becomes, you know, series advanced voice, you know, as well. I mean, I can see how the integration of contacts and your messages and things like that, I can see I could benefit there. But now, you know, now we're getting back into proper like rag implementation. Everything per user, per phone, you know, into as well. Like is that functionality gonna be a great experience, you know, and stuff there when you start to integrate, you know, things, you know, we sort of integrate things, you know, with advanced voice and real time API, you know, and then you're encompassing that on top of a rag application per device, you know, and you're still keeping your data private, you know, and stuff in as well. It's gonna be a good experience. I don't know anything. I don't, I don't know. But again, it's, it's, it's still. The infrastructure of, that is still open AI and that itself is still not the Apple way. And so, which is. I also, again, put this in question of why they're doing this. 

Adrian Green: Okay, I see. I can hear you say that, like, okay, because it doesn't seem on brand for them to, to go this direction, which, you know, I understand, but, you know, that can expand a broader topic. 

Marlon Avery: You know, really, they could be doing this to eventually bear their own. Like, okay, that could be the plate. So it's up to as well. But as of right now, it's, it's, it's a big question mark for me. 

Adrian Green: Yeah, yeah. So what about the cost? Because there's a cost reduction too. Like, um, so if it's running locally, you're not really paying open AI, you know, you won't need the pro plan. If you have something function on your device that you can. 

Marlon Avery: Is there, is that, there's a, there there's no, there's local for whisper is. I don't think there's any local for advanced voice, is there? 

Adrian Green: Oh, nothing. I know, but I know that the Apple intelligence is supposed to run locally. 

Marlon Avery: Man, they go, they still gonna figure out a way how to get paid like, you know, so. For sure. I mean, you can remember they made that, they made, they made that big announcement of that partnership deal. Yeah. 

Adrian Green: And so, but on the consumer, you know, they're not paying direct. 

Marlon Avery: Yeah, yeah, yeah. It won't be. Whoa. 

Adrian Green: Because that's what, like when what I'm experiencing locally, that's one of the best benefits is that I'm making all these queries and I know that it's not counting towards, you know, my funds and my OpenAI account. It's, you know, it's all local. 

Marlon Avery: That's interesting because what I didn't think about is that Apple could, especially when they, when they complete building their own version, they turn it into a service like a subscription, you know, just like, just like, you know, the Apple cloud iCloud. I mean. Yeah, it turns description. Yeah, okay. Yeah, I can see that. Yeah, I can see that. Okay, cool. So speaking of Apple, some former Apple employees, they left some time ago and started their own startup, a little company by the name of humane. And so we've talked about humane a few times. You know, we, I definitely have my, they could have been better, you know, conversation, but here we go. Humane on Wednesday announced $200 price cut to his flagship product, the AIPM. The Bay Bay Area startup founded by two former Apple executives, has reportedly struggled to sell the product, which led in April for 700, was another April for $700. In a message to send to his email list, humane highlights its return policy in the duPage. It says the AIP starts at 489, includes the first month of humane plan on us, the company rights along with a 90 day return window. It's complete, completely risk free unto trial. The device was poorly received by reviewers. By August, the returns of the Aipain have reportedly began to outpace the sales, leaving some 7000 to 8000 devices in the hands of users. The company reportedly began exploring Excel in May amidst poor reviews and poorer sales. Adrian, first of all, you know, I feel about them, but go ahead. 

Adrian Green: You know, the, so it's a dollar 200 price cut. So that's what from 699 to 499 for the device still requires a dollar 24 subscription fee a month. $24 a monthly subscription fee. 

Marlon Avery: Yeah. 

Adrian Green: You know, I feel like their move is going to be to sort out maybe some integrations or, you know, really piggyback on maybe some existing products out there to find a good market fit for this product. You know, we've covered a couple of these kind of everything AI products where the really, AI is great when it's directed, when it's more specific towards a certain task. 

Marlon Avery: As of today? 

Adrian Green: Yes, as of. Yeah, as of. As of today. I'm sure that's going to change in the future, but, you know, I, I feel like, you know, it's just, that has to be what they're thinking now, according to the reviews, it seems to be, and I'm talking specifically about Marquis Brownlee's review. And, you know, he. The hardware itself, how it was built, the design of it. 

Marlon Avery: My opinion, what. 

Adrian Green: I actually think he had problems with all that. I'm trying to find something nice to say about this, but I don't know if I'm gonna be able to find here. Okay, hold on. 

Marlon Avery: I got you. Hold on. I got you. So I I've saw it in person. I've experienced in person everything. I got to use it. 

Adrian Green: Okay. 

Marlon Avery: The design on the hardware side is sensational. It's a beautiful design, and it is. It sits well. It's. It's hidden enough to where, you know, people are just aren't looking at it every time they walk by you, but it also, it's, it sits well where it can be noticed, too. At the same time, it has a nice matte finish. Everything on the black, everything. The camera, the two cameras or things sits well because it has the, the cameras on the left. So it has the camera, and it has, like, the, the projector. Like the red. The red projector and stuff to it as well, where it projects, you know, images and things, the menu stuff on your hand. And then, you know, it's magnetized with the battery and stuff on, you know, on the inside. So just like, I just like this mic. 

Adrian Green: Yeah, just like your mic. I was going to say that. Yeah. 

Marlon Avery: Yeah. So. 

Adrian Green: But it's heavier than that mic, though. Yes. 

Marlon Avery: Yes. So, yes, we're still on the design, so, yes, so it's the magnets of inner. And so it sits well and may get to the point where you can almost forget it's there, you know, something there. So that's the design of it. The first issue is the battery. Yep. The battery, um, it doesn't, you know, it comes with two batteries or anything. And so, um, my buddy Matt, he has one, and he said basically, he gotta swap batteries, like, midday. And so, and he doesn't like use it a lot. And so, so that's the battery. The second, the second major issue for me was its own ended pinned it phone number. It doesn't partner with your devices. Their whole thing when, so they did a TED talk. Not yet. Tedx. They did a TEDx and stuff talking when they announced it. Their whole thing is that it was going to replace the phone. And I knew that was an immediate issue. It was immediate issue is because we don't really live in a world anymore other replacements. We live in a world now of the ecosystem and the partnership with the devices. So remember for quite some time Apple was promoting that the iPad was going to replace the laptop. 

Adrian Green: Yep. 

Marlon Avery: I have my iPad here with the notes for the show and I have my, have my, my Mac, you know, and stuff here, you know, and currently on the platform that we're using. It's, it doesn't run well on the, you know, on the iPad, you know, as well. And so they've refrained from that type of dialogue and stuff now and now is looking at it as a device that's in the Apple ecosystem and they're finding use case for specific people who potentially can use it full time, you know, so maybe, you know, a college student can use it full time and as their own standalone device, but it's not the, it's not the full replacement of a tool that, you know, we live with. So, so, well that's the, that's the next mistake is that they gave it as designated phone number and it's very, very difficult for it to incorporate, you know, with a, you know, your own devices. Like I'm not about to import a thousand contacts to this device, you know, just to have people call it on this device only, you know, as well. It's just not a good experience. And then the, the next issue of what it is, it's super laggy. It's super laggy. It's, it lags. It doesn't give you like, it doesn't give you no, no type of response. Here's, here's one of the major issues except for them. It was an on device LLM and so they were building their own, they're building their own model. It wasn't being connected and stuff. We needed models and stuff. Here's the problem with that, is if I'm on this device, I'm on this device and I have my experience with a platform like anthropic and chat wt and I am talking to advanced voice. I'm doing chat, you know, functions, everything what do you think, mentally, I'm doing in comparison? When I talk to here, I'm like, man, this is okay. Nice bet. Got that. Okay, cool. Next. All right. Yeah, I. Let me do the advanced voice. Okay. Let me do this. And I come back to the AIP and I'm like, hey, you know, I'm asking her the question and give the query, and here I am waiting. And then when it finally responds, it's nowhere near as good as models that I have access to on my phone. You're shooting yourself in your foot. You should. 

Adrian Green: And models that you have access to on your phone that are not on device. They're making a web request to get that. And it's slower on the. On device. LLM on that pin. Yeah. 

Marlon Avery: As a consumer, consciously or unconsciously, I'm doing comparisons. I'm comparing every single time I use it and everything. And anytime it doesn't match my experience or the quality that I'm seeing of, like, your competitors, it just starts to do this. And then you get to a point, you're like, yeah, I'm done with it, bro. This. Go return it. Like, you know, and everything. And so, yeah, it's not. It's not surprising at all that they have, you know, 8000, you know, whatever. So I think it was like, you know, they're getting, like, 8000, you know, returns or whatever. Maybe like, it's not surprising at all because, you know, they set a high goal. They set a high goal and the goal itself just was misaligned and they. They didn't align with the times. They were trying to step into a world that didn't exist. And that world right now doesn't exist. You know, even. Even your autonomous vehicles, your autonomous vehicle partners with. With companies and platforms. You know, your Tesla has Apple music, you know, you know, your Tesla, you know, has audible, you know, and six that you. There's no world anymore that lives where you just gonna have this device and it just lives and stuff by this own. Without the partnership of other tools, services and things like that, it just doesn't exist anymore. 

Adrian Green: Yeah, I mean, possibly the humane team just doesn't have phones. So they just weren't updated. Because if you're gonna say that this is gonna replace my phone, there's no instagram on you. There's no this. All of, you know, the creature comforts of a phone are not here. Plus, it's slower. So. 

Marlon Avery: Yeah. Can you imagine that being a policy at humane? Like, don't you dare bring your phone into the office. Microsoft employees couldn't use Apple devices. 

Adrian Green: Oh, yeah. I thought it was still like that. 

Marlon Avery: Yeah, yeah. And so, I mean, they can now and everything, but, yeah, for a time period, they can use Apple devices. So can you imagine having a policy like, don't you dare bring your phone. You figure it out. If you can't figure it out, this is why you need to get back to work. 

Adrian Green: You get in a low light area and you use this laser beam, you know, interface to get your context, you know. 

Marlon Avery: Yeah, yeah. I try to refrain from, like, promoting, you know, others, like, content, but, you know, the Marquis Brownlee review on that device was just like, I mean, he should rename that video to, like, our rip. You guys are done. 

Adrian Green: Oh, my goodness. Oh, my goodness. I think I have had to have watched that probably about three or four times completely. That was. Yeah, yeah, that was rough. That was rough. There's so many. I'm seeing a lot of. Just speaking just on that topic, a lot of creators that are kind of taking that approach, like, oh, we're going to do kind of like, expose journalism on the, on the tech industry and, you know, have these hard hitting reviews and interviews about what's going on. Coffeezilla and the other one on decoder. Decoders been releasing some content on social media where they're doing interviews and just going at. Going at these companies like it's. And with interviews with the CEO's. They did it for Rabbit and they just recently did it for Quickbooks turbo tax, I believe it was. So. Yeah. 

Marlon Avery: Hey, I'm a, I'm a, you know, speaking of policy, I'm gonna set new policy for this podcast today. What's today's date? 

Adrian Green: Today is the 23rd. 

Marlon Avery: Today, October 23, 2024. I announced today as the last day we discussed Rabbit. 

Adrian Green: Yeah. 

Marlon Avery: Like, we're just gonna stop. Like, I don't. Somebody. Yeah, I don't. 

Adrian Green: I don't want humane me. That may be that way, too, man. We, you know, maybe this can be the day that we sunset the humane discussion, too. Unless something really changes, you know? 

Marlon Avery: Humane really built something, though. Like, humane. They really built. They really took some time to think and innovate. They just. They just missed the mark. They missed. They missed, like, I would argue they missed one crucial mark one day, which is the integration of your phone, you know, the integration everything. And it being. It being a learning device and stuff with your experience. So basically it should become a wearable, like the Apple Watch everything. That's the. That's what they should have, like, and I believe they would have succeeded. Everything which also, too, this gives a company like Apple, like we just talked about, it gives them roadmap to go build a device like this in, like, a few months or a couple of years or so, what it might be, but they innovated and they built, like, some really nice hardware and ip, and they built, you know, LLMs and stuff in integrating stuff with it. And, you know, they figure out a way how to give a device that small its own, you know, phone capability. You know, rabbit, again, is just a raspberry PI, bro. Yeah, it's not that raspberry Rabbit is basically the hacker version of something, like. 

Adrian Green: Humane, but it seems like in that hacker version, what were they, 70% to 80%, possibly, of the features that the humane paint the pen has besides the hardware. Take the hardware out of it and take the question answering take. Take the actual functionality of it. It's the same complaints for both of them, you know. 

Marlon Avery: Yeah. 

Adrian Green: And I would say that, you know, with. I'm not even going to defend these guys. I'm not gonna. But, you know, with rabbit, you know, it's very hacked together. It's very, you know, hacked together on the back end. 

Marlon Avery: Right. 

Adrian Green: And, you know, throughout, you know, whatever they did with that, they basically got the same reviews. I mean, they. It was the same problems. There was battery life complaint. There was also the slowness complaint as well, which, you know, they fessed up and said, it's not an on device LLM, I think, which, you know, with the humane pen, I'm kind of thinking maybe the same. The same situation, because it's so. It's that slow. And I was just telling you about what an on device. How fast an on device LLM is. Yeah, it's way better. 

Marlon Avery: Yeah. Yeah. Oh, also, didn't Tesla. Didn't they fess up and said that the Tesla robot at the. At the announcement that basically it was being controlled by humans? 

Adrian Green: Yeah. Oh, wait. Control by humans as far as, like, the voice? 

Marlon Avery: Like the voice. 

Adrian Green: Oh, the voice. Okay. 

Marlon Avery: Yeah. 

Adrian Green: Okay. 

Marlon Avery: Humans would basically talking to a mic and everything, instead of it being, like, our own device, you know, on the own device LLM, you know, and talk and talking experience in that. 

Adrian Green: Interesting, interesting. I'm still. I'm impressed that that wasn't. 

Marlon Avery: It's difficult. 

Adrian Green: Was that the extent of the. It being driven by a human? Because if these robots were walking around this crowd, you know, serving orders maybe, and making cocktails, was. How was that someone back there with a remote control kind of doing that? Or was that. 

Marlon Avery: If. Here's the thing. If if you put one thing in question, you gotta put everything in question. 

Adrian Green: Yeah, yeah, that's true. 

Marlon Avery: You gotta put, you know, you gotta put everything in question. Everything. And so they wasn't, it wasn't forefront, they wasn't forefront in being transparent. They basically just allowed your imagination to roll. 

Adrian Green: Yeah. 

Marlon Avery: And stuff. And so, and they could have just came out and said like, you know, these are our human power, you know, AI, you know, Tesla robots. 

Adrian Green: Yeah. You know, because that's, that's still a story. Tell me that you got a bunch of ready player one people in the back with headsets on direction, you know what I mean? And then actually, that's amazing. That would be great. You know, someone operating an avatar, basically sign me up. 

Marlon Avery: What a job. That's, that's hilarious. All right, man. Okay, here we go. Adrissin Horowitz has now announced a new program called the Oxygen program. The big venture capital firm is now providing access to Nvidia H 100. GPU's a key resource to AI startups for developing their large models. This is significant because the tech giant like Google Meta, Microsoft have been dominating the GPU market, leaving smaller companies struggling to secure computing power they need to compete. By offering flexible and low cost access to GPU's, a 16 Z is giving their portfolio companies a chance to grow without the burden of expensive cloud contracts or inflated evaluations needed to afford these resources. In broader context of ADA development, ASICs and Z is positioned themselves as a key player, not just through financial investment, but offering the essential infrastructure for AI startups to need to survive. The GPU clusters is a game changer, is leveling the playing field for smaller players in the race. It allows their startup to focus on building constraints of fundraising for computing costs alone. Adrian I starts with this. So as you guys know, I used to work at a venture capital firm a couple years ago and I was engineering director there for a couple years and a venture capital firm was known to be a full stack venture capital firm, which means I, that they wrote the check, but they also dove into the operations side of, you know, with you. And so they were right to check, but they help you with legal, they help you with engineering, they will help you design, you know, your content and things like that. And, and so it gave us the opportunity to thrive in different ways and partner, you know, very uniquely with, you know, entrepreneurs who need it, you know, more than just cash, you know, so you can get cash or anything. But, you know, even right now it still still has some complex of like hiring people, you know, at scale, this is brilliant. I mean, there's no other way to put this. This is, this is, this is brilliant for them to acquire, get access to um, a bunch of, or maybe even front the costs. So maybe they, they front the calls and they kind of include that, you know, within their, you know, how they set their valuation, their name to the city startups. And so I think this means also, I mean they're, they're funding so many AI stars right now. I mean it's, yeah, it's, it's crazy. And so, but no, this is brilliant. And also too, if you are a, if you are a startup and you're looking to, you know, get some, get some VC funding or, you know, even it gets some VC funding, you know, I think it's gonna be very, it makes them, makes them set apart. I'm not sure if, I'm not sure if any other VC firms or stuff are doing this. I haven't heard you're saying, I mean, but they're definitely large ones who can, you know, gain access to anything to bust GPU's or either front the cost and stuff like that. But I mean, they're also ahead of the market. I mean, they're ahead of the curve, you know, stuff. And so, you know, if you are a startup and you're looking, you know, you need for, you know, definitely, you know, you could use to kind of build out your infrastructure or scale, I think it's hard to ignore that and I think it's hard to ignore that. And I think it's just like a smart move, you know, from, from the team. And so for, was to, you know, to be able to do so. And so, you know, now, now you give a startup the ability, you know, to build a, a model for a particular use case and not get over charged, you know, from, you know, your loud cloud, you know, cloud infrastructure companies. And so I think we talked about, I don't know if we, I'm not gonna say it cause I can't remember if we talked about this privately or we talked about on the podcast, but, you know, we, we know there's a cloud company who has figured out ways how to decrease their in house costs for GPU's and inference, but they have not changed their consumer pricing, which, I mean, there's nothing illegal, anything to that less overall. Yeah, and so there's nothing legal to that. But you know, it's, it's just kind of funny because I would argue they're probably going to increase their price, you know, too, as well. You know, on the cloud side, you know, for the AI compute and stuff. And so, yeah, and then, and then also too, we, I know this, we didn't talk about publicly. We also know a cloud of the cloud provider who is struggling to get GPU's a large one. And, you know, it's starting to come out now that one of the competitors have all the GPU's. And so a lot of companies are trying to, you know, switch ship, you know, jump ship and stuff over there. And so, yeah, I mean, being, being able to kind of like not be in between at war and have your investor partner, you know, kind of like front the bill for you, if you will, is a nice thing. It's like that. I think my only question or can maybe concern stuff here is does the, does the valuation of those companies kind of remain the same? Do they increase, do they decrease anything? Can they use something? Can it use like that type of like asset only through a cap table, you know, since they did get an investment or anything from, you know, dreason Horowitz, you know, and also too, does it give companies like injured capital firms like it Vitrisa horrors, did they give them more power over the startups if they need to do a takeover, you know, if they need to switch seats, you know, they need to switch seats or thing, you know, with CEO's or for employees, does it give them more access to power over these, you know, these little startups? So, I don't know, but yeah, I think it's, I think it's overall good, but definitely if, with everything there's definitely has some consistent. 

Adrian Green: Yep. It's a, it's a strategic move. It's a strategic move like an accelerator, like, you know, y combinator or, and you're seeing horowinds. I don't know if they're an accelerator, they're more of a venture capital firm. 

Marlon Avery: But yeah, venture capital for a y combinator is the accelerator. 

Adrian Green: Is the accelerator. Okay, so, you know, they're gonna, it's, they're hopping on the way, on the way on what's popular right now. So before it could have been credits to cloud infrastructure is what they would give you if you, if you, you got investment. 

Marlon Avery: Ah, that's right, yeah, good point. Yeah. 

Adrian Green: Before, you know, an access to certain things, so, because that is what's in demand right now as far as the companies that are of getting the funding that are getting picking up steam, they have to do with AI. So what this is going to do is it's going to attract people to get investment from Andrasene Horowitz and Andreessen Horowitz so that they can hop on that wave and they'll, it could potentially be mutually beneficial, because if this company, let's say company does something that actually catches, then the venture capital firm benefits as well. So it just positions them great in the market as far as demand. Because right now, the story of the startup is not the same as it was ten years ago. The whole game has changed. The whole game has changed. It was a game of, you know, build fast, launch a product, you know, buy some ads, get some steam, get a user base. Now it is a lot of these, a lot of these things have been really, I don't know either. What's the word I'm looking for? It's not really as easy as it, as it was before because the market is flooded really right now with a bunch of people with applications. And there was a time where you can launch an app, but then it'll change your whole life. I don't think that we're necessarily in that economy anymore. So, you know, really, I would be interested to know as far as what the competitive edge this is going to give this venture capital firm in five to ten years. I would, I would, I would be interested because things change so fast. I, things change so fast. And what they're doing right now is smart because they're hopping on the wave to attract those companies that are going to need that back end. But they're doing nothing different than what they've always done and to what I assume they've always done, which is to have basically the infrastructure to support these startups. So, yeah, it's definitely genius. It's definitely a good move, but it's, it's biz business as usual to a certain. 

Marlon Avery: Yeah, so, yes, this, this gives them a strong, this gives them a Runway. This gives them a Runway to be wrong a lot and find those unicorns. And so this gives them the opportunity to invest wide and also deep, you know, so typically, so typically, either they have venture capital firms, you know, are set up, is like you have a fund your day. So say, for example, your fund is $100 million. And let's say, for example, you focus on three categories, you know, let's say, you know, e commerce, energy, you know, and, you know, hardware, whatever. So you, you take a certain percentage, anything, and you allocate that percentage, any for internal, you know, user things. So pay your employees or they, so typically that's, you know, typically that's like below 5% everything for like the, you know, general, but there's ways to increase that. And so then the, you supposed to allocate, you know, that 90, 95% everything to, you know, the, you know, the funds or things, different companies. And so you put that in three categories, you know, so and so. But with that, you know, those categories now become limited. So you can't necessarily, you're not going to invest, you know, if you're writing check size, you know, up to like, let's say for your series B type of company. So you're writing checks out somewhere. Coming back can be like, you know, five to 25 million type of thing. And so now you're not going to really, you're not going to invest that into a lot of companies because you don't have like the, the financial upside to do so. Now you're still kind of limited with that. What this does with them is it allows them to write smaller checks. Oh, for sure, at a wider range, you know, and so now they can write smaller checks because, hey, we're funding the calls, you know, for, you know, your GPU's, you know, access to all these things. And so it's like you said to as well, you know, typically one of. 

Adrian Green: The things, I mean, they're, no, they're, they're renting the GPU. So I mean, I thought that, you know, they give them money and then they would be paying as they go either monthly or something, a rate to, to use those GPU's. 

Marlon Avery: Yeah, so, like, so, yes, yes. So, like, like either way they're renting. They acquire a bunch of, you know, GPU's and stick. It's like basically they're still kind of fronting the cost of it because then if you, if you don't have this program and then you got to go get into that cloud war, you know, you know, you know, with these large companies as well. And so that's going to be a higher cost, you know, to as well. We know with that. So you can say cloud war. You get into that cloud war if you can't do this, anything. So now this allows you to write smaller checks and invest. You know, why? And now since you have, since you now have wrote in smaller checks or anything, you still can create. So you can still can have a certain percentage where you can double down on the ones that you notice really about to explode. And so, you know, you, you can find those unicorns very quickly when you have that type of infrastructure. And so if you develop into, if you invest into 100 companies, you're probably gonna have like one or two unicorns. And now you double down on those two. And so now you can write, you know, now you go to, you, you initiate this deceive the seed fund near name that, you go to a series a, and then now you can double down like on a series b and stuff with them. And so, yeah, it's a, it's a, if it's an infrastructure play everything, it's not a new thing earning. And the only new variable in this is the GPU's. And also to dresen horrors, does the same thing with, with talent as well. And so a lot of people don't know, they invest into companies to get access to talent. And sometimes, you know, they will go snatch the talent from one company and they go place them and stuff in another company. Yeah, that's really like kind of, you know, taking off. And so they do the same thing. Could they do the same thing too on the legal side too, as well? So they got lawyers. If you ever been to San Francisco, the Adriatician Horowitz office is like surrounded by, like, they got like three or four lawyers. And so, like their prone personal lawyers are like next door walking distance, you know, and stuff, like right there with it. And so they use these same individuals for, you know, their portfolio companies when something, you know, big and stuff happens. And so this is why this, this whole notion, this kind of goes to something else. This whole notion that the Bay Area is done, it's going to go away. It's things like this why they won't, you know. And so when you have, when you have, number one, you have an incubator, which is waste combinator. And so you're finding the early companies who can potentially go, and then you make the another investment into them when they come over to a system z, you have an infrastructure around you with like your legal team and stuff with it. By the way, you're down the street from Stanford, you know, as well. So you now you're having recycled talent. Not recycled, but you're having new blood Thailand coming in every single year, and you're just finding stuff, people. And so, and then you also still have, you're right down the street from downtown San Francisco where talent can be laid off at any time, you know. So now the lead engineer from Snapchat just got laid off. Yeah, come down street. We got a company for you. 

Adrian Green: Yeah. 

Marlon Avery: You know, stuff as well. Like that bear is never going anywhere like anything. And so because they, they develop an ecosystem and community that will thrive past anything. 

Adrian Green: Yeah, I'm with that. But I have a little bit of doom and a little bit of potential doom and gloom to interject here. But about this model. So they're renting the GPU's to the startups which to me is sounding like Monsanto. It's sounding like you know, kind of like a record industry, you know, where there's an advance that you have to pay back. So if their business model is that we're going to invest into your company, however as you're renting the deep to GPU's, this is, you're paying back a certain amount every time like or every month depending on how you're running the DPU. So you know, you have this cost that is taken away or you know, deducted from your you know, overall investment every, every month as well. And you know depending on the, how that's priced you know, androsine, andra scene, Horowitz can be making money off of that in and of itself as well. So if they're, I mean if they're renting, okay, watch it. 

Marlon Avery: I don't think so. I don't think they're, I don't think they're charging number one. No, no one is illegal to do so from the VC standpoint. You can't, you can't charge your, you can't charge your startups for services. Um, it's very very heavy. It's very very, a lot of red tape and stuff there. So you can't, you can't. 

Adrian Green: What is, what is renting a GPU then like what is what is written? 

Marlon Avery: So watch this. You said it earlier is basically you say it like hey, it's like a loan and stuff almost, if you will watch this attrition horrors are not writing grants. The investment itself is a loan without the GPU's. So they're just including this in the already system alone. So this, this 2 million, 10 million, whatever else if I'm giving you is included in now in the GPU, you know, so it's already a loan. Like they're already giving you loans if they want to invest into you. They're not grants. And so they're just including this now this new program saying like hey by the way when you take our investment now you have access to this oxygen program that allows you to get direct access to h, it's included in the loan. So again it goes back to it, now it goes back to is, does the valuation change for that company? Does the valuation change for that company? Because now this is almost a a second loan stream from the VC to the startup. I mean, it's all bundled together and stuff with that. But now, does this give them access to more power? Like, does it get them a new board seat, you know, over the company, you know, now, does the 25% of the company now become 35% just because of this commodity is so needed stuff right now in this. In this world? I don't know. You know, I don't know. But, you know, you can't, for VC's, you can't charge your startup, like, services, service fees, anything that's like. That's like, holly. It's a bunch of red tape and stuff in that area, you know? Yeah. And so, but you can. You can't include, because what? It's a full stop venture capital firm. So, like I said, we did the same thing, a light ship. So, like, we can't. We couldn't charge them when we help them with the design, you know, of their website or product, we just include that on the initial investment. 

Adrian Green: Gotcha. Okay. That makes sense then, because. Yeah, and thanks for. For clearing that up because, yeah, I wasn't sure about what that. What that model looks like, but now. Yeah, now I know it still counts. 

Marlon Avery: On drawbacks for sure. I mean, I'm not. I'm not over. Not over stepping it. Definitely still have some drawbacks. And so we'll. We'll. We'll see here then. We'll see what happens. Up with that? 

Adrian Green: Yep. 

Marlon Avery: Well, great, man. Um, as always, my brother, man, this is. This has been great. You know, we. Yeah, man. Oh, by the way, too, man. We. We miss Sekou Seku. Everything had to step aside. He has some, uh. Uh, some. Some business things take care of, and so, uh, hopefully, he'll be back with us and stuff next week and everything. And so psycho manner you're listening to as well, man. You know, uh, love you, miss you, and they looking forward to come back. We. We held some conversations back because you weren't here. So that was our dedicated dedication. Yeah. So you. And so, uh, so, yeah, man, we talked about the. This AI Photoshop debate. You know, we discuss anthropics new model or thing that now can have control over your PC. Apple intelligence is about to roll out, as well. A humane price drop. You know, they dropped $200 on a price which is now started starting to send more red flags to humane. And then Adrisse Nahorwest has made a power play surrounding the GPU's, and so, yeah, man, it's been a great episode. 

Adrian Green: Agreed. Great episode. Great talking to you, Marlon. 

Marlon Avery: Well, guys, definitely, like, comment, subscribe, definitely. You know, give us a, give us a like and stuff and subscribe. So to the channel, man. We are here weekly at every night, 09:00 p.m. on Wednesday we live stream here on Twitch, along with YouTube as well. And, and then now you can find us on all platforms at AI with Friends podcast. So, guys, until next time, we are out. Peace.