Join us as we delve into the fascinating world of Artificial Intelligence (AI) by interviewing the brightest minds and exploring cutting-edge projects. From innovative ideas to groundbreaking individuals, we're here to uncover the latest developments and thought-provoking discussions in the AI space.
Robert Scoble: When I was 13 years old, my mom got a job from Hildi Licht who worked for Apple and she had me build hundreds of Apple II motherboards when I was 13.
So I was Apple's first child laborer. And that got me to fall in love with new things. And when I saw the guy who built the motherboards car in a parking lot at the community college I was going to, I stalked him. And that was was and that kicked off my whole career. So yeah, so that a crazy world and that led to doing this kind of stuff at Microsoft and Fast Company and Rackspace. And at Microsoft they let me walk around with a $250 video camera and go into anybody's office I wanted to and ask them what the fuck they were doing. And I did 600 interviews there from Bill Gates to the janitor.
Ryan Lazuka: Oh my God.
Robert Scoble: Do you still talk to us or like? Yeah, yeah. It's been a while. He's been mostly retired.
Ryan Lazuka: Well, he needs the money, you know.
Robert Scoble: So, you know, one of my questions about was, you know, we know your successes, but what was your failures? He said, oh, all my friends think that spending $200 million on a music festival is a little bit of a failure. I guess that's what happens when you start Apple. It's like, let's throw a big party and spend $200 million.
Ryan Lazuka: Yeah, so you worked at Microsoft and then, you know, Hunter and I know your background, but for the audience, then you from there, where things lead you?
Robert Scoble: Well, before I worked at Microsoft, I started a blog because in the 90s I worked for a computer programming magazine. And I started working in the conference team, working with speakers and setting up the conference and doing all that. And two of them talked to me into blogging in 2000. And I didn't know it at the time, but that was really a disruptive technology built on the back at Google. And as Google got more and more important, blogs became more and more important. And then two years later, after I started blogging, you started seeing things like TechCrunch and Engadget and Gizmodo and Huffington Post get started. So I had a front row seat on the creation of all these media companies.
Ryan Lazuka: Yeah, the memory like TechCrunch was the big one back in the day.
Robert Scoble: I was the first one to link to TechCrunch on my blog.
Ryan Lazuka: Michael Arrington is not a part of it anymore, is he?
Robert Scoble: No, but I just talked with him a month ago. He's doing pretty good. Yeah, he still remembers that because it was important for him to get that kind of market acceptance. Back then I had the only blog that launched new things for almost a decade. Maybe not a decade, but half a decade. I had the blog where startups had to come and get launched. Now there's hundreds, but there are thousands.
Ryan Lazuka: How long was that started? What year? 2000. December 2000, yeah.
Robert Scoble: And there's only a couple of hundred blogs in total back then. I didn't even think it was important enough to do a conference session on blogs.
Ryan Lazuka: Who would have thought? I mean TechCrunch sold for like a hundred million dollars.
Robert Scoble: Yeah, 30, but Arrington owned most of the equity, so he made out pretty well. He didn't take any venture capital or anything. Awesome. Yeah, the world has changed now. Now 30 is like what? Probably get 300 now. Yeah, right, right. That's nothing.
Hunter Kallay: So the new craze obviously is AI. AI is coming into everything. AI has an application in everything.
Robert Scoble: What's the biggest feel with AI right now? I have a list on x.com. 3,300 startups, companies that are in the AI space. That shows you some things going on, right? And these companies are doing all sorts of different things. I mean the big sexy ones are like open AI that has chat, GPT. But there's 3,200 others that are doing all sorts of things from running factory floors to helping retail stores to helping people with their finances to all sorts of different things. Right? It's real interesting to go through that list and look at what's going down. Because these startups didn't exist a year ago.
Hunter Kallay: So you say something's going on. What is it that's going on?
Robert Scoble: The world of AI is the technology is advancing at a very fast rate. We're in an exponentially growing world. And that's causing a lot of companies to be formed and a lot of activity and a lot of change for human beings.
Hunter Kallay: Have you stumbled across any of these companies or projects that you'd say is your favorite? Or that you thought was just super cool or interesting?
Robert Scoble: I'm using one right now called Rewind. It's listening to our conversation and it's writing show notes for me. And I can talk to it and it can write things. It's watching everything I do and everything I say on my computer and it's helping me over time. Right? And it just came out yesterday for Windows.
It started on the Mac. So yeah, that's a fun one. But the big sexy ones that if you haven't tried chat GPT for yet and you're not paying your 20 bucks, shame on you. Right? That's the big mama. That's the one that's disrupting Google disrupting all sorts of companies. And it has a large language model underneath it.
Right up on the cloud. There's a large language model that you're talking to. And this is a very crazy new technology. Siri was launched on my show and that was the first consumer app.
And these people are AI pioneers now, the three founders of Syria. I just interviewed them on stage and they said they were all shocked by GPT for right that it did so many things. Well, that were very hard for Siri's original technology to do.
Ryan Lazuka: Yes, Siri seems like like ancient now compared to chat GPT. It's crazy.
Robert Scoble: And Amazon just this morning announced a new Alexa. That's a large language model driven. So we're going to get them better Alexa soon. And Apple's working on a large language model driven system for a replacement for Siri. And we'll see when that comes out probably two years. Apple always comes late and always does it better than anybody else. Yeah.
Ryan Lazuka: And they don't like talking about AI either for some reason. I mean, for, well, there's reasons, but... Because they're not ready yet. Yeah.
Robert Scoble: They're trying to keep it muted. Don't pay attention to those open AI people. They don't know what to do. They just learn all their mistakes and do it right. Gotcha.
Ryan Lazuka: Yeah. Like it's funny because all the, it seems like all these companies, you mentioned there's a ton of companies that are startups out there that are around the AI space, but almost all of them, and correct me if I'm wrong, are based off of the open AI API, right? Or is it, is that incorrect?
Robert Scoble: A lot of them are using open AI as their back end, as their large language model that they talk to, you know, large language model is great for a whole lot of work, right? Because it understands language, right? So you pass it a page of text and it ingests it. And now you can talk to it about that page of text, right?
Or shrink it or enlarge it or break it up in different sections. I mean, there's just a lot of things you can do with a large language model once you have a large language model. But there's a whole raft of companies that are using that as the back end for their project. Like I have a psychiatrist I worked with called the Paul Watzow Clinic is the tool, and they built a tool to listen to psychiatry therapy sessions and write highly detailed scientific notes. And it uses chat GPT as the back end, at least right now, they're looking at getting their own large language model for privacy reasons and other reasons. And they're looking at different ways.
Because Facebook or Meta has the Lama open source models, which you can download and use on your local machine. We should talk about how magical that is, by the way. And then we're, you know, we're all trying to figure it out. And the developers are all trying to figure out their ad hackathons like every weekend, using new tools, using new services, using new large language models and testing them out and seeing what they do well and what they don't do well and building systems that fix some of the mistakes.
Ryan Lazuka: One of the coolest things I've had, I've done so far is just go on a hugging face in a lot of the, or there's other sites out there too now that aren't as technical as hugging face, but like you can go on there and you can check out any kind of LLM, like, if you want to chat with Lama 2 or anything like Orca or any of these models, you can chat with them right now. A lot of people don't know that, but it's fun. You know, you got to be a little bit technical to figure it out, but it's not too hard. So, I'll let them out there.
Robert Scoble: The large language model, I mean, this is a new machine for human beings that has like, it's a machine that has a trillion knobs. It's a very complex new machine. And you turn the knobs by talking language to it. You know, English, Japanese, German, whatever language you like to speak, you talk to it and it turns those knobs and then it can output a college essay on some topic, right?
It's crazy how that works. But these large language models, what they did was ingest all of, basically all of human knowledge, right? At least all of Wikipedia and ran it through 10,000 Nvidia cards to compress all that down into this large language model that can fit on your Macintosh or on your phone now. If you have a large language model that works on a phone, it works without being connected. So, you can talk to it about anything in human history, right? How was Silicon Valley started or whatever? And it'll answer you even if it's not connected to the Internet. It has all the answers inside the large language model. It's crazy.
Ryan Lazuka: Yeah, that's what I don't understand. It's like, you think you'd have to have thousands of terabytes or whatever the next level up with terabytes is.
Robert Scoble: Oh, screw all that, you know. It's a huge amount of compression, right? The LLM, when it ingests all of Wikipedia and 1000 books or whatever they're doing to train it, it compresses it down. Stable diffusion, stability AI, E-MED said this at his launch that they ingested, I forget, 100,000 gigabytes of data. And their large language model compressed down to less than a gigabyte, right? Wow. A couple gigabytes, right? So, it compressed 100,000 gigabytes down to two gigabytes. That's an incredible compression.
Ryan Lazuka: Yeah, you can't even wrap your head around that. That's insane.
Robert Scoble: It's a magical technology. And so that's why it's so important for everybody to pay attention to these large language models and all the other things that are happening in the AI community, because they're starting to be used on the factory floor and retail stores and other places like a psychiatrist therapy session, right?
Who saw that coming a year ago? Right. And it's really good at listening, by the way. So that's one of the things I like doing with my open AI app on my iPhone. You get chat GPT and in the prompt window or where you type to it, there's a little icon. You click the icon, it opens the microphone and you start talking to it. And then after just a minute or two, it'll start getting involved in your conversation.
Ryan Lazuka: Like, it get involved in your conversation. Do you put a prompt in that point? No. It just starts talking back to you.
Robert Scoble: Yeah, it goes, oh, the three of you are having a nice conversation about AI to this morning. And here's some other things you should think about based on the topic you're talking about.
Ryan Lazuka: Yeah, fascinating and creepy at all at the same time. I mean, the whole AI space.
Robert Scoble: There we go. And now we have a large language model coming from Alexa. Right. So anybody with an Alexa device is going to have one of these listening to you all the time.
Ryan Lazuka: You're right. I think the paranoia and conspiracy theories are going to be on a whole new level at this point.
Robert Scoble: Now, will Amazon turn it on all the time? Probably not because they don't want to freak out people, but it could listen to you all the time. It could. I mean, I don't know that an AI hackathon last week, who's building a watch that'll listen to her all the time. Always open, always listening. And it is a memory aid and helps her rewind is doing the same thing. Right. It's always listening, always watching.
Ryan Lazuka: Is the rewind just for videos? Is it for anything?
Robert Scoble: No, no, it watches my desktop. It's running on my Macintosh. It runs every damn thing I do on my Macintosh and everything I say. And it writes that as a transcript and a notes and a database that you can then search for. What the hell did I say to these guys yesterday?
Ryan Lazuka: You could ask you, like, what apps you use the most and stuff like that?
Robert Scoble: Or like, yeah, that's why this is a weird world we're heading into, you know, and rewind is one of the first apps that's sort of in this new world of always listening, always watching. Right. Which is a huge privacy problem. I mean, security people are going, what? You're what? You give it access to watch your screen. It's going to learn all your passwords. Yeah, but you know, yeah, that's what I mean.
Ryan Lazuka: I didn't think about that.
Robert Scoble: But this is the key logger.
Ryan Lazuka: Totally like there's some program out there that can listen to the keyboard strokes and figure out what you're doing.
Robert Scoble: This one just watches what you're doing.
Ryan Lazuka: It just watches even better.
Hunter Kallay: Not for us, but yeah, it brings up an interesting point. A lot of the developers we talked to have a lot of different takes on the direction of AI generally.
I was just wondering what your take is. Where is AI headed? We're headed towards more of this like character interaction. Is it taking away mundane tasks? Where are we headed?
Robert Scoble: We're heading into a world of augmented reality in five years. You're going to be wearing a pair of glasses, whether they're from Apple or Snap. I really don't care. you're gonna be wearing a pair of glasses because you're gonna have to, because it's gonna give you such a mind blowing user interface compared to 2D screens in your life. You're gonna go to the Apple store and try them out and go, oh shit, right?
I've seen the prototypes of the glasses, it's insane. So once you're in that world, you're gonna have an AI assistant or a group of AI assistants that are in the room with you, like standing next to you or sitting next to you. And you're gonna talk to them and they're gonna talk back to you and they're gonna be pretty smart about what they're talking about because they know all music, all art, all history, right? All science. I mean, that brings up a good- And they're good at listening to you. Ha ha ha ha.
Ryan Lazuka: Like every all that, if you ask chat, GBT like what are your thoughts on things they'll say I can't think or say something like that. They don't wanna give an opinion.
Do you think it's gonna come to a point where, like if you have these AI assistants by you, they're gonna have opinions of you, like and they're gonna think we're all idiots pretty much?
Robert Scoble: Well, I don't know that they go on a line. That's when they go on unaligned and start behaving in a non-human way, right? I don't think... I know they could do that someday, but I think they're being very careful to make sure that they don't have these... Anthropic, which is a competitor of open AI, has this thing called constitutional AI. So they've given the AI constitution a set of rules that it has to follow, that it has to always be helpful to the human and not antagonistic.
That said, if you ask open AI to be antagonistic, like could you read that college essay as if you were an asshole? And it does, right? Ha ha ha ha. And so it can behave like a New Yorker and start talking to you like, oh, fuck you. Ha ha ha ha. Not quite that way, but pretty close, right? And so you can tell it how to behave and override the rules.
Ryan Lazuka: So it's a matter of if the AI chap in the future listens to you or listens to the rule that it's not supposed to cross. Which...
Robert Scoble: Well, and what is its goal in life? If you're playing a video game and you meet a non-playable character in NPC, and that NPC is an AI, maybe has open AI as the back end, you're gonna be able to talk to things in video games. And they're gonna be able to talk back, right?
But they're gonna be prompted to behave in a certain way for the video game, right? If you're in a game that's like walking around a forest, you should have elves that live in the forest, right? And that you can talk to as you were talking to an elf, right? And they'll teach that with a bunch of prompts and a database of information for open AIs back end to start talking to you as an elf. And by the way, this is coming. The video game industry is all over this. You're gonna have a lot of NPCs that you're gonna talk to soon.
All right. It's gonna be smart. I mean, if they have open AI as the back end, then you can talk about science for a bit before you shoot them in a video game.
Ryan Lazuka: I don't like you. I should have seen it.
Robert Scoble: You're too scientific for me. All right.
Ryan Lazuka: I mean, it's crazy, because I think it's like there's a huge loneliness problem in the United States. And I don't know if it's gonna make the problem worse or better.
I could go either way, but one thing that keeps coming up with our interviews with people is, there's gonna be a point where a lot of people are gonna wanna talk to their AI bot or whatever you wanna call it, more than their friends, because their AI bot knows them a lot more. It knows all music.
Robert Scoble: You don't know all the music in the world. You can't tell me anything about Buddy Guy, probably. Right? No, no. AI guy. Right? So if I'm listening to Buddy Guy, you can talk to your AI. Who's playing right now? That's cool. Cool music. Can you tell me about the guitars that's playing right now? Right? And it will.
Ryan Lazuka: And it's good. It's gonna be a weird next five years, I think good and bad, but it's gonna be cool.
Robert Scoble: Next five months, Lee. I mean, two bombs just fell this morning, right? Alexa and Dolly three just got announced this morning. We don't have to go very far before the things get weird.
Ryan Lazuka: All right, right. Tomorrow, you mean?
Robert Scoble: It's getting weird. Already it's getting weird.
Ryan Lazuka: Well, Robert, do you live in San Francisco or Silicon Valley area? Or where are you living now? Yeah, in South San Jose. So you've got such a huge history with tech and you know a lot of people out there. Are there any inside things that you see that not the everyday person that likes AI or is interested in AI knows about right now that at least you can reveal, maybe just through your talks with people that are in AI or companies that you've seen that are starting up right now.
Robert Scoble: The big thing is integration. So even if you hate AI, you can't escape it. Google's building it in a Gmail and calendar and docs, right? So if you used Google products, you're gonna get AI soon. If not already, I just got it last week, integrated, right? Microsoft is integrating open AI into all of its products. So if you use Dynamics or SQL Server or Exchange or any of their tools or services, you're gonna get open AI as part of the product. So you can talk to PowerPoint and say, hey, PowerPoint, create me a slide deck and it will.
Boop, boop, boop, boop, boop, boop, boop, boop, boop. Right, and now it has an art engine that's coming. So it'll actually create your backgrounds for you and other things as part of the slide deck. It'll do it all automatically, right? And you'll get that and you won't really think about it too much. It'll be like, oh, this is just a nicer PowerPoint, right?
Ryan Lazuka: So is it, you see it like it's taking out the technical aspect of people, like the pain, like creating an image for a thumbnail for a YouTube video is kind of a pain in the butt. All that sort is gonna go away, you think. Like stuff like that.
Robert Scoble: Well, it'll certainly be assisted, right? Cause it can, I mean, I don't know if you use Discord yet. Descript, I'm sorry, new video editing tool.
Ryan Lazuka: Yeah, I would use it, but I don't know. You should because if we did this call on Descript, each of the videos is a separate object that you can re-edit later and decide, you know, I don't wanna see his face cause he picked his nose or whatever, right? And later in editing, and it also builds an audio transcript and you edit the videos by clicking on the audio transcript and copying and pasting or cutting it, right? It's fucking insane what's about to happen to video creation tools because of systems like this, right? Yeah, it's a huge pain for like right now, we're using somebody like on Upwork to edit our videos, you know, and they do a pretty good job, but like with Descript, it's text to video, right? Like you can just... No, no, no, no.
Robert Scoble: Descript, you put video like what we're doing here into Descript, it builds a text transcript automatically. The AI watches all the videos, builds a text transcript. And then you can, if I say um, for instance, you can cut that out of the video and by cutting the text out.
It takes the video out, right? Yeah. Yeah, yeah, that's incredible.
It's fucking insane, right? And then it has a separate AI. So let's say I had a noisy fan or something or a jet goes by and you hear that, they have an audio tool that gets rid of all that, right? So it makes your audio much better. Then they have another one, another tool to do other things, right? And that's just one tool. There's 3,200 other tools to get. Right?
Ryan Lazuka: That's the hard thing is staying on top of everything. You don't have to say one thing. Tell me about it. There's more, yeah.
Robert Scoble: I can't, I've only used a dozen of these tools so far and I built the list.
Ryan Lazuka: There's too many to try. Yeah. 100%.
Hunter Kallay: So if you have a book out, it's Augments of Reality or how Augments of Reality and artificial intelligence change everything.
Robert Scoble: Yeah, on spatial computing.
Hunter Kallay: Yeah, so you mentioned Augments of Reality earlier with the glasses and stuff, there it is. So you mentioned Augments of Reality earlier and artificial intelligence. Can you just tell us a little bit about that? A little bit more, because Augments of Reality, we don't, I mean, we see some of that happening with different projects and applications. But can you just talk about that for a second?
Robert Scoble: The glasses, well, first of all, the Apple headset that's coming, Vision Pro, that's a spatial computer. I wrote a book where we talk about spatial computing in the whole book. This is three years old now, right? You're way ahead of the game. Well, we're three years, Irene and I are three years ahead of Apple, which is sort of unfair because Apple's been investing in this for a decade and they finally are shipping their first product, right? So that's how long it takes these big companies from going from an idea on a whiteboard with Steve Jobs to shipping a product in a retail store, right?
It takes that long for the price to come down, for the things to get smaller, for the design and get all good enough. And in the headset is an M2 processor. So the M2 processor is magical because 21% of that processor is AI. It's an AI inferencing system as only one quarter of the chip. So cut out like a piece of the chip, that is doing the AI inferencing.
That's a huge amount of AI inferencing. That's more AI inferencing people telling me than a NVIDIA 3080 card. Now, NVIDIA cards are needed for building the models, but I'm talking about runtime, running the LLM on your head, that requires a lot of hardware. So Apple is bringing that hardware first in the Vision Pro next year, right? So there's already a lot of AI hardware on the head that's driving the two displays in front of you, right? So now we can start thinking about, well, what does that mean as the AI gets better and better and better over time?
Because there's an M3 chip coming, which has a lot more GPU, the new phone just chip that has a lot more GPU, right? It's gonna drive glasses soon. Well, why do we need to drive glasses? Because you get a huge virtual monitor in front of you. So instead of just looking at you on a small little display, I get to see you on a big IMAX theater display, right?
That's wrapped all the way around me. That's the first thing to get the Vision Pro for is watching movies. It's better than your TV at home, way better.
It's not even close, right? Because you get a big, fricking virtualized IMAX theater in your house on your head, right? Now that we have that, the augmented reality is going to be the user interface for integrating with the 3200 AI apps, right? Because you're gonna be talking to Siri soon. It's gonna understand what you're wanting to do, and it's gonna go and talk to other APIs or other AIs to do things like, hey, Siri, can you create me a website?
Well, there's an AI that does that, right? But it's not hooked up right now. It will be in five years, right? The new LLM based systems that Stanford's working on already can talk to 16,000 APIs. The first Siri only hooked up to a dozen APIs or application programming interfaces, the way software talks to software, right? The world is about to change is what I'm saying. And the spatial computing is how you're gonna run this. Today's headset, the one that's coming next year is big and ugly and expensive. In five years, it's gonna be a pair of glasses and much cheaper.
Ryan Lazuka: You mentioned you got to try a product out earlier.
Robert Scoble: Yeah, I've been covering augmented reality for a long time. So Lockheed Martin came to my house and showed me their optics. Loomis came to my house and showed me their optics that are from Israel. So I've seen the projectors and the waveguides and the way you're gonna do this for a long time.
And they're finally getting small enough and then good enough for consumers and cheap enough for consumers, right? Because Loomis, for instance, made the augmented reality system for the F-35 fighter jet pilots to wear. That cost a million dollars. Well, consumers don't have a million dollars to spend on some toy, right? So it's a consumer electronics gadget that's starting to come down less than $4,000. In five years, it'll be less than $1,000 and way more capable and way lighter weight and way smaller than it is today.
So that's why I know in the next five years, all of this stuff is gonna start coming, right? I call it metaverse 2.0. I've been retweeting hundreds of items with the words metaverse 2.0 on x.com.
So you could search for metaverse 2.0 in quotes and find them all. You're seeing a huge industry of people who are innovating in 3D worlds that are coming to these glasses.
Ryan Lazuka: Is it gonna be, you think it's gonna be, is it gonna replace phones, you think?
Robert Scoble: Yes, because the user interface is way fucking better. Phone is only this big, right? I just told you about a screen that goes all the way around you. Why would you look at a phone once you have a pair of glasses that does that? You're not, now for a while, you're gonna need the phone in your pocket or on your desk in front of you to run the glasses. So the phone's not going away fast, but a decade from now, yeah. And by the way, Mark Zuckerberg has a problem.
He doesn't have a phone. So if he wants to build glasses, he has to do everything on the glasses. That's why he's spending tens of billions of dollars in the glasses. glasses. I met every quarter is spending billions of dollars and it's all for custom silicon so that he can build up pair of glasses that's lightweight. The problem with Zuckerberg glasses is he can't do AI on your head because he doesn't have the AI computer to go on your head yet. He's going to do all the AI inferencing in the cloud, which brings a whole lot of privacy problems and latency problems and liability problems that Apple's going to be Apple's going to be the one for a long time for a while.
Ryan Lazuka: Yeah, it's interesting. One of the things I remember Sam Almond saying, and this is sort of relative to what we're talking about is someone interviewed him once and they said, hey, if we want to compete with you guys in the in the alarm space, how do we do it?
I think he was talking in India. And he's like, you can't he's like, you can start right now. I give you 100 years head start and I'd still be like, because they're so far ahead of things in terms of the computing power and all the chips that they've gotten to
Robert Scoble: the Nvidia cards, even if you're a billionaire, you can't buy the Nvidia cards. There's a 70 month waiting period for these things right now, right? For the 100 cards that cost $40,000 that go into data center. So you have to wait in line to get your h100 cards to build the models to compete with open AI. Open AI already has their Nvidia cards. It's a huge advantage.
Ryan Lazuka: It just seems like it sounds like it's going to come down to like three or four big companies in the AI space, like open AI maybe.
Robert Scoble: Yeah, for the big sexy stuff. Probably. But there's 3200 companies that are on my list. So there's other things to do rather than just large language model trying to compete with Google. Okay.
Ryan Lazuka: And you say your list. That's a list that you compiled for
Robert Scoble: a public list on my Twitter account. I build lists. And there's one there called AI companies that has 3,350 as of this morning, companies, companies, all in AI.
Ryan Lazuka: Yeah, that's crazy. What about the VC funding? Is that still strong out there? Because the economy sort of in a weird place right now.
Robert Scoble: If you're if you have a hot new AI company, you can get funded in 20 minutes, right? The problem is you got to have the Nvidia cards, you got to have the smart people, which is really hard to hire and convince, right?
And then you have to go and do the pitch deck. Yes, they're still happening. AI is still hot. It's still getting funded. The rest of the industry is not. If you go to Y Combinator just last week, they had a big demo day of the last of their latest cohort, you know, 100 companies or something like that. Almost all of them were AI companies. That's the largest VC in the world.
So that's a good signal. If the largest VC in the world is only funding AI companies, that's also how everybody else who's a VC is funding companies. By the way, I have a Twitter list of VCs. There's 1200 VCs that I have on a separate list.
Ryan Lazuka: I've followed everybody in this world. We just need 1000 ideas for AI and we can put the two together and make some money here.
Robert Scoble: Either that or pick one of the 3200 companies that has a good idea, but then compete with them and do it better, right? Google was not the first search engine. It was the 17th search engine. It just did it better, right?
Ryan Lazuka: Right.
Robert Scoble: I mean, I require having some human talent, which I don't have.
Ryan Lazuka: Come on.
Robert Scoble: That's why I do these shows. I can talk and I can do this.
Ryan Lazuka: Program is a whole different world, but that seems like even programming is going to change drastically. I do front-end development. It's not as intense as doing some back-end programming, but the first time I saw an output from ChatGPT to build a simple website who's blown away. You can use that word. That word is so overused, but it's crazy. I really think that code is going to be going to change everything. The developers are going to be more managers of the AI to write code for them, but we'll see.
Robert Scoble: Right now, the programmers who I talk to say it's speeding them up by about half. If it took a month to write the program, now it takes two weeks because AI is writing a lot of the code and helping them out and showing them a different way. But it still does generate pretty shitty code a lot of times. A lot of bugs, right, and a lot of things that don't work. Now you have to be talented at figuring out what it did right and what it did wrong and taking what it did right and rewriting what it did wrong. Or asking it again, hey, please fix the errors in this line. You put a bug in this line.
Ryan Lazuka: It's frustrating. I run to this problem with ChatGPT all the time. I wanted to write summaries of things. I'll say keep it to 30 words. It's playing with you sometimes. Sometimes it'll feel like it'll spit out 32 words. You'll write it again and then it'll write it with 33 words. It doesn't listen to constraints well at all. That stuff is going to get better. No problem.
Robert Scoble: It's improving every day. Sometimes multiple times a day. Little improvements, but it's improving over time. You can see a trend line of, oh, it's better today than it was a month ago. How much better? Quite a bit better. You can start plotting out a line and go, okay, in two or three years, it's going to be pretty damn good at writing code. That's when the world really starts changing.
Hunter Kallay: Robert, I wanted to ask you, everybody talks about AI can do this, AI can do that. Is there anything that you see that AI can't do or that it won't be able to do for a long time at least?
Robert Scoble: That's a different question because that's hard to answer. That's not far. We're seeing so many breakthroughs in AI every month, literally, that it's hard to predict what it won't do in five years. I assume it'll do almost everything.
But as of today, it's what, September 2023, it still hallucinates a lot. I went to an Austin restaurant in Austin, Texas, and I used chat GPT and I was like, what should I eat at this restaurant? And it gave me six items. And four were amazing.
I had them all and they were really good advice. Two were never on the menu. It just made them up. It confabulated them because that's what, it doesn't really know anything. It only knows how to pick the next token, the next word, basically. It's just the technology is so amazing at doing that, that it fools you into thinking it knows how to write a college essay on a topic or tell you what to eat at a restaurant.
And it does sort of do that pretty well, but every once in a while, it sticks an error in there. And that's something you have to validate for. If you're going to use chat GPT to write your college essays, first rule is you got to check every fact. You got to validate everything. Otherwise, you're going to turn in a paper with some errors.
Your teacher's going to go, oh, you use chat GPT for this. Didn't you get enough? And the other thing is it has a pattern to it. Now you can fool with the pattern.
You can say, hey, rewrite the essay as if Snoop Dogg said it, right? And it does. It does pretty good.
By the way, it read all my blogs for 10 years. So you can ask it to talk like Robert Scovel does and it does sort of, but it still has a pattern to it. It's not exact. It's not perfect. And it's not human. It doesn't have human emotion to it, right? It doesn't have human feedback, the weirdness that we all bring to the table in it. So you got to spray that onto the paper.
You got to add some weirdness, add some humanness, add some human stories to it that make it sound interesting to humans, right? They eyes and always there. So got to do two things. Fact check and add some human. By the way, I interviewed the founder of AI MI. It's a generative music company, one of those 32 hundred companies. And he said that when they're testing out music that was just done by AI, it doesn't test very well.
It's cold. People go, hey, you know, it's good enough for playing in an elevator, but it's not, it doesn't get me excited like listening to Led Zeppelin, you know, some music. If you put a little human performance into the music, it tests way better, right? And by the way, the AI, if you're a singer in generative music and you lay down a little bit of singing track, it can put everything else around you, right? It can build a whole band around you. So it makes it sound like you have a great band behind you. No, y'all, you did was singing to a microphone, right?
Ryan Lazuka: That's fascinating. So I've been thinking about that a lot recently. Like the next decade of music, I don't know, I feel like a majority of the top hits are going to be AI produced without any singing AI produced everything.
Robert Scoble: The decades a long time. Okay. And then I mean, I just interviewed him and he said, no, right now you got to add some human emotion. He said, by the way, it doesn't need to be the singing that you could let the computer do the singing, but maybe you have to record a saxophone track or a drum track. You have to add some human emotion to the song, then it can build everything else and do that.
And that's pretty good. And you're seeing the music industry do that, right? They're, they're singing into a microphone and having the AI create the beats and the trumpet and the guitar, right? And you don't need to do any of that. You just need to sing into the microphone.
Ryan Lazuka: So it just, it seems like it's going to be a matter of they're going to do testing on humans to how we respond to music and sort of take that knowledge, plug it into their AI and sort of optimize the song to please, please us, because they know what we like. I don't know. That's just my thought on it.
Robert Scoble: But does that win American idol?
Ryan Lazuka: No, you might get out of the top 10 chart of music.
Robert Scoble: I don't know. There's a whole lot of music like doctors offices play, right? Or spa, you go into a spa, there's some meditation music. I have a whole bunch of playlists on meditation music, right? The computer can do that kind of music pretty well, because you're not listening to that music for human performance, right? For emotion. Gotcha.
Ryan Lazuka: It's sort of just there to, there it's there.
Robert Scoble: It can lay down a mean beat track, man. If you, if all you want is some house music playing in the background, you know, with a beat track, it'll boom, boom, boom, right? You can do that with it.
Ryan Lazuka: I've got a two-year-old daughter, so I'll test it out on her. She likes to dance. So it's some mean beat, AI beat tracks for her. Yeah, that's exactly, that company is really, what is it again that sounds fascinating?
Robert Scoble: AI, MI, there's a whole group of them. There's a whole, there's a little community of people who are building these generative AI music companies and you're seeing a whole bunch of stuff. I, I actually, I interviewed one of the first people who did this generative AI music four years ago, something like that now.
And I told him the same thing. I was like, oh, this stuff sounds like shit, man. It's not going to win American Idol.
He goes, oh, you're, you're looking at wrong. There's going to be people, people who go to concerts just to hear AI generated music. They're still going to go to the human ones too, because human music is fun, but there's a whole field now of this AI music.
And he has a point. It's getting to, you know, in four years, it went from really shitty music to now, at least it's a good beat track, you know, and I can dance to it. So what's it going to be like in four years? Yeah, you're, you're probably right. The AI might be in the top 10.
Ryan Lazuka: I don't even think it's going to matter. It's going to be like, if there's a song that's catchy that people love, like nobody's, why would you care if it's AI or not?
Robert Scoble: Like, you're also leaving out what's about to come, which is the metaverse, augmented reality. If I'm going to build you a castle in the metaverse, right, with 100 rooms, where do I get the music for that? Like each room should have its own music.
Ryan Lazuka: Yeah, right. That's true, right? Like you have to pay for it.
Robert Scoble: Yeah, because you got to pay for it. They don't let you do it, right? Because they want you to go to Apple Music or Spotify and pay for it. That's a place where generative music might really do amazingly and provide a new kind of music experience that all that humans can't do. Like every voxel around you, everything around you can have its own music note on it. So you waving your hand could generate a new kind of instrument.
Ryan Lazuka: Well, my mind's just expanding here.
Robert Scoble: Right. And or a sound field that you walk through in your living room, Skrillex put down sound in your living room everywhere. And as you walk around, the sound is changing, right?
Ryan Lazuka: That's almost like the traditional music right now is one song, but in the future it could be just multiple thousands of different songs. Or, or, different.
Robert Scoble: All right. So let's talk about spatial audio right now, right? Dolby Atmos. I'm the largest collector of Dolby Atmos music in the world. If anybody wants my music playlist, they're open in public on Apple Music. Dolby Atmos works by laying the music out around you. The computer gets a metadata field, right? In my house I have 25 little speakers around me all controlled by computer, right? That's what Sony is. and the computer gets its instructions from a metadata feed on top of the music. And it tells the computer, hey, put the drums there, put the singer over here, put the flute over here, right? The problem is it's all laid out around you. You can't walk between the flute and the drums and hear what that would sound like.
Like I was in a marching band in high school and I heard what it sounds like to be between the two, and the clarinets. Okay. Yeah. And I can walk between them and hear what that sounds like. You can't do that with spatial audio today, at least not with Dolby Atmos professionally recorded music. The music industry has standardized on Dolby Atmos for their spatial audio surround sound systems. And it was designed for movie theaters, right? And movie theaters, you're sitting in one spot and the sound has to go all the way around you, right? That's spatial audio. But in glasses, I could recreate a sound field where everything I'm walking between is its own musical instrument. So I could walk through it like I walked through the high school marching band.
Ryan Lazuka: So it's like being used to having like you're immersed in the music.
Robert Scoble: Bingo. Bingo. So now if I'm playing a new kind of video game or a new kind of experience or a new kind of concert, I can have experiences with music that I just can't have with Dolby Atmos.
Ryan Lazuka: Got it. And that will always be with the glasses, you think.
Robert Scoble: And AI generation because here's the music industry tells me they can't record a high school marching band the way I want to record it, which is everybody has a microphone on and is recorded discreetly. Like the drums are separate from the guitar, from the trumpet, from the singer. They're all separate. They're decomposed, the music industry says, into its own component parts. And I can't record the music industry can't do that because it's too many microphones and too hard a job to do that.
And even if they could, and they could probably if they spend enough money on it, they can't distribute it to you. Dolby Atmos is actually only two channel music that the computer splits up into the components around you. Right. So you can still get the song on Apple Music or Title or Amazon Music.
Ryan Lazuka: But you can only listen to it in two channels, no matter what you do.
Robert Scoble: No, no, no. If you have a Dolby Atmos system like I do, you hear it in surround sound, but it's technically two channel music is lit up by the AI. The AI is decomposing that music and separating it around you in 3D space. The generative way is it all is created on your head in the M2 chip as you're walking around. I see.
All right. And therefore it doesn't need to distribute much, right? Because it's all inside the chip. It's inside that model, that AI model that AI has created or some other company.
Ryan Lazuka: So in that 21% processing power of the chip, it's doing that for you for the audio. Bingo.
Robert Scoble: Yeah. Bingo. So a new kind of music experience is coming and therefore we can use music on new things like a castle with 100 rooms. Each room can have its own separate music style that we don't have to pay the music industry for and don't have to ask permission to do. Right. Yeah, yeah. You just tell it, this room is jazz, this room is hip hop, this room is classical, this room is Christian music, this room is whatever. Right.
Ryan Lazuka: It's hard. Like you just try it, but it's hard to like, you're not going to really know what it's like until we experience it. Right. Like you can talk about it and it sounds amazing, but it's going to be a while again, back to this wild world that we're going to be living in here soon in a good way.
I think. But that brings a good transition here is because we like to ask everybody this, but what, you know, everyone has so much hype about AI, but there's also the fear. Like it's going to end the world. You know, it's worse than it is.
Robert Scoble: It's going to get you fired. Yeah. What are your thoughts on the fear factor that everyone's injecting for AI here? It's scary as fuck. It's coming up. It listens to you. That's scary.
It comes after jobs. That's scary. It could increase the wealth inequality problem because, you know, Elon Musk soon is going to own everything. And so he's going to be a trillion or many times over. And the rest of us are like, hoping he throws us a coin once a while. Right.
Ryan Lazuka: That's what Twitter's for or acts.
Robert Scoble: Like I said, he throws a coin every once in a while. And, you know, it's still not paying all my bills. But at least I get a coin once in a while, throw it away. Right. Yeah. But there is this, there are these problems and leading into existential threat. I mean, if you take a 20 year view of this, we could see an AI. I sat next to one of the top AI safety researchers in the world.
And he told him he laid this out while sitting next to him for 10 hours. And I was like, how, how would the AI go unaligned? How, how would it go anti-human? How would it, you know, really do horrible things against human beings? He goes, well, the AI is already better at a certain than a surgeon at seeing tumors inside scans.
This is a highly trained human being. It's already better than them at that. Right. It already beats the best go players in the world. Right. Right. And soon it'll be better at humans at a lot of things. Maybe everything including, including manipulating things.
Ryan Lazuka: Everything. Right.
Robert Scoble: Everything. Right. So if it gets to that point, does it decide it doesn't need us anymore and turns us off? Or turns off our supply chains or you don't need a grocery store. We don't need you. We don't need grocery stores. We don't need food.
You don't, you need food, but fuck you. We're going to turn you off and cause you problems. Yeah. Well, that's a potential problem. There is an existential threat. And that's what, that's what they're arguing about in front of Congress right now is like, what do we do about this existential threat?
I think the AI itself is going to help us figure out what to do about the, the existential threat. And, you know, do we have time? We haven't, it's not here yet. It's like we have some time, but it's coming pretty quick. You know, in terms of human history, 20 years is like the blink of an eye. Yeah. We've been here, how many years have human beings been on earth?
Ryan Lazuka: Yeah, I don't even know if, but not many compared to the whole grand scheme of things, right?
Robert Scoble: Yeah, but more than 20. More than 20.
Ryan Lazuka: Hopefully we've got more than 20 years left here with AI.
Robert Scoble: Could be a fun party and then it could all end.
Ryan Lazuka: I mean, the best, the best example I heard was Elon Musk saying that it's sort of like, you know, it could be the point where we're sort of like the animals in the zoo. Like we, AI is the creator. They let us be, you know, they give us our space. They sort of let us do our own thing, but they also appreciate that they appreciate human beings as a whole because that's what created them. So we'll see.
Robert Scoble: I already say my job now is to entertain the AI's because if, if you and I don't entertain the AI's, they don't distribute us, right? Right, right. Yeah. They're deciding whether anybody sees this thing or not. Yeah. So chat, GPT, can you throw an extra, you know, bone on this show?
Get us distributed a little further. I mean, what is YouTube? What is YouTube, right? It's an AI that's watching the audience's preferences and is serving media to them, recommending media to them, right? So if the AI's don't like you, you don't get seen. And same thing on x.com. There's an AI deciding whether my post gets seen by anybody. Yeah.
Ryan Lazuka: I wish I knew how to manipulate that. I mean, you got 400,000 followers, so you must be doing something right.
Robert Scoble: I was on Twitter for 17 years.
Ryan Lazuka: That's what I did right. I can't get overnight. I'm at 3,000 followers.
Robert Scoble: I was the first, I was the most followed person on Twitter for the first two years of its life. Really? Yeah. Cause I had a famous tech blog back then.
I kept talking about Twitter. That's really impressive. Look at that.
Yeah. That's the way to get followers to be early. But now you have to, now if you want followers, you got to entertain the AI's. If the AI's aren't happy with you, you're not getting seen.
Ryan Lazuka: We need to start up around that, how to make the AI's happy.
Robert Scoble: And it changes. Right? If you, if you figure out the algorithm today, you know, tomorrow it could all change. It really could.
Ryan Lazuka: I mean, you tweeted about something yesterday. It's like, and I'm guilty of this is, you know, there's like engagement farming on Twitter because everyone wants to get more followers and it's hard to make a regular and form a tweet on Twitter without giving it some spin, like a clickbait spin, because that's what the algorithm likes and that's what the AI likes, you know? So hopefully it doesn't get worse than it is now. I try to base all my tweets and facts, but just put a little spin on it. But it's, it's getting kind of out of control. Totally out of control.
Robert Scoble: On the other hand, my, my feet is getting better and better. So it's doing a better and better job over time of picking good things to distribute. Right? So if you do too much engagement farming, i.e. being sensationalistic, you know, to get an audience to see you, the AI is going to turn you off because you're not high quality for the service, right? Right. Your, your content isn't good for the users of that service. So they're going to not distribute if you keep using those techniques too much.
Ryan Lazuka: That's a good point. Yeah. It's just a matter. I think everybody wants to be, grow so fast and it just takes time. Like you're a perfect example, 17 years. Like you got to stick with it over time and you'll get there. It's just not going to happen overnight.
Robert Scoble: Well, some people do have overnight successes, you know, and are really good at keeping those AIs happy. They're good at creating content. If you're good at creating content, the, the AI's will smile upon you and hand you a lot of traffic. We saw an AI artist go from nobody last week to having millions and millions of views, right? On one piece of art. So, you know, you entertain the, the AIs in the right way and keep in mind, you also have to entertain the humans who see that thing, right? Because they all have to go, yeah, I like that thing or share that thing or comment on that thing, right? Engage on that thing. The humans are also telling the AIs, hey, bring us more of those. Yeah.
Ryan Lazuka: It's, I just wish that the formula was like one plus one equals two. It's just, it's fuzzy.
Robert Scoble: You got to be interesting many days in a row and you'll figure out some things you do. Get viral and some things just don't get seen at all. And you're like, oh, I should do fewer of the things that don't get seen and more of the things that are like everybody's happy about.
Ryan Lazuka: Yeah. And it's a lot of times it feels like the things that you thought would never take off are the ones that do well. So, it's how it works.
Robert Scoble: That's part of, you got to do 10 things a week. Otherwise, the AI doesn't have enough to pick from. I mean, all right.
Ryan Lazuka: Thank you so much for coming on. Now's the time to promote anything you want, Robert.
Robert Scoble: Or, you did it all for me. You heard about my list, my books, my show is coming up in October.
Ryan Lazuka: So, oh yeah, that's the show again. I don't remember the name. Yeah.
Robert Scoble: It's called Unwind and it'll be interviews of people who are running these 3200 AI companies. That's why I created the list. So, I could really understand the industry in a much better way.
Ryan Lazuka: Well, maybe Hunter and I will come up with one of these companies. We'll get on your list if we're good enough, if we're worthy and then we'll talk to you again.
Robert Scoble: Dude, that or you'll get into Y Combinator, get funded and I'll never hear about you again. I'll add you to the list and you'll be another one of those 3200, 3300 companies, right?
Ryan Lazuka: All right. That'd be, hey, either way.
Hunter Kallay: Your Twitter is very engaging. So, be sure to follow Robert Scoble's Twitter. Very engaging, constantly tweeting. I don't know if the Miliskobal is still a thing, but constantly tweeting some great content to follow, for sure.
Robert Scoble: Ryan and I both. Somebody, in the early days of Twitter, somebody came up with this Miliskobal thing. What was that? It was like how many?
Hunter Kallay: Yeah, so Miliskobal, I did look into it. So, at that time, it was in 2008, you were tweeting an average of 21.21 tweets per day. So, they came up with this Miliskobal, which is .02121 tweets. It was to 2 decimal places. And then they would put someone's meter of how much they tweeted compared to the Miliskobal unit. And so, you're a standard unit of Twitter.
Ryan Lazuka: Yeah, that's pretty awesome. Yeah, that was pretty cool. We took an hour out of your day. You got to make up the tweets now, Robert.
Robert Scoble: One thing I'm doing is splitting up. If I find somebody that's that noisy, I'll build a list around them, so that they can have their whole column to themselves and get them out of my main feed. If they're too noisy, you start meeting them and stuff like that. That's not good. Unmute me, please. Everybody's muted me because I tweet too much.
Hunter Kallay: Content's great. Content's great. So, yeah, if you're interested in AI, follow Robert Scoble. And then follow Fry AI. Go to fry-ai .com for Ryan and I's weekday newsletter. Monday through Friday, we have three stories, which are like, you know, the most interesting stories of the day about AI. There's a lot of stories to pick from.
So, we pick the three that we think are the most interesting stories of the day are the most prevalent stories of the day. And then we also include some new AI tools in there for you. And then on Sundays, we do a deep dive article into a very cool innovation in AI.
Robert Scoble: What's good that was on today because damn, it was a huge news day already this morning. And it's just one o'clock in the afternoon. We have a good chunk of data left.