This might be the coolest thing you see today. This big fluffy purple guy that you see right here, his name is Hugo. And no, that's not somebody in a costume. And it's also not AI. It's actually a twolegged like humanoid robot under all that fur. He's from a company called Dynamic Creatures. And they built him to walk around places like theme parks and hotels and just hang out and interact with people and kids. Here we can see them walking around their office, giving high fives, popping up from behind computer screens for some reason. Now, one interesting thing about this company is that the guy who runs it used to be the chief strategy officer at Boston Dynamics, which is a company known for making robots that are pretty impressive and could probably actually mess you up. Hugo's got 57 joints and 43 actuators. And in this really cool moment, this little girl runs straight at him and he actually opens his arms to give her a hug. Now, right now, there actually is a human controlling him remotely. So, this isn't like fully autonomous yet, but they are planning to test these out with real customers starting in 2027. Personally, I think it's kind of cool, and I can definitely imagine this type of thing walking around a place like Disneyland. What? I'm curious. Would you let one of your kids go up and hug this
You've probably seen them in your feed. Huge mustaches, wild hair, dead serious faces. They're not real, but they get millions of views. >> [music] >> One page has 12 videos and about 83 million views. So, why is everybody making these and who's paying for it? Behind that page is a man with a huge blonde bob, a curled mustache, and a brown suit, and he's far from the only one. A bald man with an enormous mustache, a tie, and yellow pants has a single video at 52 million views. And a guy with a wildly inflated body and hair hanging over his eyes has more than 90 million across the 30 posts I could see. I kept running into pages like these, and at some point I got tired of just scrolling past. So, I'm running an experiment. Build my own AI influencer, launch her, and show you everything from the first drop-down menu to the first real numbers. But, before building anything, I wanted to understand who actually pays for all those views. The short answer is brands. Companies want people to see their product, and they'd rather pay creators for views than buy ads nobody watches. Right now, one way to get a piece of that money is paid creator campaigns. Brands post them on creator platforms, and as an example, I'm going to open one called Launch Point. [music] It's a marketplace where creators browse campaigns, pick one, make videos, and get paid. When I open the explore tab, it's basically a long list of brands that want videos made. There are AI video tools, prediction platform, and other startups. Payouts on some of these go up to 10 or 15,000 dollars per video. And an AI character is just a very effective way to produce the attention they're paying for. The principle is simple. They pay for eligible views, meaning the ones that count under that campaign's rules. So, just as an example, say a campaign pays $1 per thousand eligible views. If a video gets 20 million of them, that's $20,000 before costs and taxes. That's if it gets that many, and most videos obviously won't. One more thing shaped how I approached the whole experiment. The money in these campaigns comes from a series, not from one lucky video. A page that posts regularly with the same recognizable character is what builds an audience. So, I needed a character people would remember and a way to make videos with her fast. This video is sponsored by Higgsfield, and that's where I'm going to build the character. So, let's do that next. The first thing to figure out is the look. A page like this lives on a recognizable silhouette, one or two memorable details on the face or in the hair, and a signature outfit the character always wears. Look back at the examples. The mustache is taken, the bob is taken, the inflated body is taken, and every one of them is a man. So, I went the other way. My character is a young woman who looks like she stepped out of a 17th century portrait, a slightly odd aristocrat, but she behaves like any influencer today. She takes selfies, does trending dances, and shows off cash. The whole joke is the gap between how she looks and what she's doing. Her name is Lucrezia Valmont. To build her, I'm going to the top menu in Higgsfield and opening the AI Influencer section, then the Create Influencer tab. [music] You can upload your own photo as a starting point, but that's optional, and I'm building from scratch. And here's what I liked right away. There's no prompt. You don't write anything. You build the character by picking parameters. There are a lot of them, and they cover pretty much everything from the body and the face to the hair and the clothes, down to small things like accessories, plus a few extra lists of distinctive details on top of that. Just the hairstyle list has 26 options, and quite a few of them are pretty strange, which is actually what you want when you're building a character people should recognize in a second. The most important one at the top is character type. It's basically how exaggerated the character is. Average is realistic, bold is a noticeable caricature, and extreme is almost a cartoon. For Lucrezia, I picked bold because she should still look like a real person in a real photo, just slightly off. An extreme would turn her whole face into a cartoon. So, here's what I picked: female, adult, tan skin, tall and slim. For the face, a long face shape, hooded eyes, and brown for the eye color. I didn't pick a hairstyle at all and left that part to the model. For the style, I didn't choose anything from the list or add any accessories. Instead, I attached a picture of the exact outfit she should wear in the style tab, pearls in the earrings included, because for this character, the outfit is half the concept. At the bottom, there's a batch size setting, which is how many variations you get per run, a switch for free generations, there are five of them, a random button, and the generate button, which costs about one credit. So, you can build a character like this without paying anything. I hit generate, and this is what came out. She has short reddish chestnut hair with a thick, straight fringe, and the ends on both sides curl sharply upward, so you'd recognize her even from a tiny thumbnail. Her eyelids are half-closed, and she has a slightly haughty look, like she's mildly unimpressed with you. She's wearing a dark burgundy velvet dress with a narrow waist, wide sleeves, lace cuffs, and gold embroidery, plus a pearl necklace with a ruby pendant and heavy earrings. Her face is only slightly exaggerated, and the hair and the dress do most of the work. Now, picture that face completely serious, filming a selfie or dancing to a pop track. Next, she needs to move. To make the videos, I'm using a tool inside Higgs Field called Genjutsu. You take a clip that has already gone viral and swap the person in it for your own character. The scene and the movement come straight from the original. Dances work, and so do reactions, interviews, memes, and famous performances. The clip has already shown that people want to watch it, so the only thing you're really changing is who's in it. Step one is finding a source clip. I went with an Instagram video that has more than 200,000 likes. A guy in a gray t-shirt and a chain is counting stacks of cash. It's simple and visual. And here's a tip worth stealing. Pick clips that have already blown up, which is also what Higgsfield recommends. Then I opened Genjutsu, which is in the same top menu. Inside, I click create video and choose the object swap mode. There are two upload fields. The first one says, "Add a reference video to edit." And that's where the money counting clip goes. It can be anywhere from 4 to 30 seconds long. The second field is for your characters, products, or clothes. I upload the picture of Lucrezia that I made in the influencer studio. You can add up to 30 images in there. So, you could also add the reference of her burgundy dress or product you want her holding in the shot. There's also a prompt field, but it's switched off by default. And for a video like this, you don't need it at all. The clip already has the scene and the motion, so the model repeats the movements on its own and keeps Lucrezia exactly as she is. For settings, I picked 1080p and left the format on auto. You don't set the length here because the result comes out as long as the source clip. The generate button shows 280 credits for this one and I hit it. And here's the result. Lucrezia is standing in the same room with the same stacks of cash doing the same movements as the guy in the original. The fringe and the hair curled up at the sides are still there. So, are the burgundy velvet dress with the lace cuffs and the pearls with the ruby pendant. >> [music] >> The camera is low and looking up, just like in the original clip. A 17th century aristocrat counting cash with a bored face actually looks natural in the shot. If you put the two clips side by side, you can see exactly what was kept and what was swapped. The motion is the same, the person is new. Now the series. I took two more clips and ran them through the same steps. Source clip, character, generate, result. The first is a viral street dance. Lucrezia does a trending dance in her velvet cord dress and pearls with a completely unbothered face. The second is a street selfie. She sets down the camera, steps back, fixes her hair, and poses. I wanted to see whether she works outside of dances, too. What I noticed is how fast it goes once you have the routine. You get the same character in any scene, and each new video takes a few minutes. I made a brand new Instagram account for Lucrezia Valmont and posted three videos there. The money counting, the street dance, and the selfie. And by new, I mean completely empty. Zero followers, zero posts, no history at all. So, nobody had any reason to see her. The money counting video is already at about 3,000 views, and it's still climbing. The dance is at around 600, and the selfie at 125. Across the three videos, that's close to 4,000 views and eight followers on a page that didn't exist before. Those aren't viral numbers, and I'm not going to pretend they are, but for a start from absolute zero, I think they're good. People who'd never heard of her stopped on her videos, liked them, left comments, and a few of them even decided to follow. A brand new account also needs some time to warm up. As far as I understand it, Instagram first has to figure out who to show your videos to, and it learns that as you keep posting. So, three videos and a fresh page aren't really the test. The real test is what happens after weeks of posting regularly. So, the plan is to keep posting and be pickier about the source clips. Not just anything that went viral at some point, but the ones that are blowing up right now. If Lucrezia shows up in a trend while people are still watching and sharing it, her version has a much better chance of getting pushed along with it. That's what I'm focusing on next, and I expect the numbers to go up from here. What surprises me most is how little it takes to get started. One character, a few viral clips, and a few minutes per video. Whether Lucrezia turns into a page that actually earns is still an open question, and 1 week is too short to answer it. So, I'm going to keep posting and find out. The links to Hicksfield are in the description. See you in the next one.
changes to Chat GPT plans. So, remember when a month ago they essentially said that they're having the usage you get with your chat subscriptions? Well, they just backpaddled on that in a big way because they're actually doubling your usage on certain subscriptions. I mean, check this out. Tibo here from OpenAI, who is becoming the communicator for these kind of things, put it out there that if you're in the 100, $200, or $500 plan, you're actually going to be getting double the usage effective today instead of cutting them in half. So, in other words, the higher up plans are getting more value while in more and more ways the $20 plans are starting to feel limited. You don't have dots. You don't have this big usage limits. But to be fair, with the value and the intelligence that AI can provide, it's I mean, $20 is an incredible steal. And we're probably entering into this era where if you want to use it for serious tasks, you'll probably have to be on the $100 plan. We'll keep an eye on that though.
Could not fetch transcript (it might be disabled or unavailable). Error: Could not retrieve a transcript for the video https://www.youtube.com/watch?v=VaNWoD7JORo! This is most likely caused by: Subtitles are disabled for this video If you are sure that the described cause is not responsible for this error and that a transcript should be retrievable, please create an issue at https://github.com/jdepoix/youtube-transcript-api/issues. Please add which version of youtube_transcript_api you are using and provide the information needed to replicate the error. Also make sure that there are no open issues which already describe your problem!
Stick figure videos are still winning on YouTube and it's not slowing down. This channel has only 26 videos and one of them got a million views in just 2 weeks and this one only 20 videos, 156,000 subscribers and one video with 17 million views in just 2 months. My last stick man video showed you how Claude makes one of these from a single command and at the end I promised to take this further. So this time Claude made a full stick man video from just one command using fewer tokens than ever. Watch this. >> Thousands of years ago, people lived in groups. Have you ever wondered why they never left? Meet Oga. He is hungry and the grassy land around him is wide alone. Oga searches for berries all day, but he finds almost nothing to eat. >> That's just a few seconds of it. The story, seven scenes, one steady narrator and every line matched to the picture. And look at the bottom corner, no watermark. But here's the crazy part. The moment Claude clicks start, it stopped working and the video kept making itself. That's because this time Claude isn't working alone. It hands the heavy work to a custom tool inside Google Flow. That's why it uses so few tokens and that tool doesn't exist in Flow. I built it. At the end of this video, I'll show you how to build the exact same one with just one paste. But first, there's a problem I hit while testing this that ruined my first videos. Every single scene said the exact same line. I'll show you exactly how I fixed it because if you skip that part, your video will break the same way. Here's all you need, the Claude desktop app and a Google AI plan with access to Google Flow. Flow uses credits for the video clips. This one cost me about 100. Open the Claude desktop app and start a new Claude code session. It has its own built-in browser, so Claude can open websites, click and type just like you. Type open Google Flow in the built-in browser pane and there it is, right next to the chat. Sign in to your Google account yourself, just once. Claude won't type your password for you. Now, the command. At the top, write your topic as a question. Mine is, "Why did ancient humans always stay in groups?" Under it, the number of scenes, seven. Then, paste my command. Hit enter. Now, look at my mouse cursor. I'm not touching anything. First, Claude writes the story, not a list of scenes, a real story. A small character, a problem, something that goes wrong, and a clear answer at the end. But, this time, there's one new rule in my command, and it's the reason the voice and the picture never drift apart. Every narration line has to describe exactly what its scene shows. So, the voice can never talk about one thing while the picture shows another. Then, Claude picks the narrator. It looks at Flow's voice list, and for a history topic, it chooses a calm, informative voice, Sharon. Now, watch the Flow window. Claude opens my project, finds my pinned tool, clears the box, and pastes the whole story in one go. And now, the problem I told you about. In my first tests, every scene used the same voice, Karen, and the final video sounded perfect for about 8 seconds. >> Hi there. We're introducing a new feature. >> Then, the narrator said the first line again, and again, and again. The whole video was one sentence on repeat. Here's why. When every scene used the same voice name, Flow kept handing back the same recording. So, here's the fix. For every scene, Claude opens Flow's voice picker, types that scene's line, and saves it as a brand new voice, Sharon Group S1, S2, S3. A new name means a fresh recording every single time. And the tool double-checks Claude's work. If two scenes ever get the same voice clip, it locks the start button and tells you exactly which scenes have the problem. So, you never waste credits on a broken video. Now, something happened in the middle of my run that I didn't plan for, and how Claude handled it surprised me. Flow's login expired, and the voice picker stopped opening. Most tools would keep clicking and make a mess. Claude didn't. It stopped and told me exactly what was wrong. I reloaded the page, told Claude to continue, and it picked up right where it left off. When all seven scenes have their voice, Claude clicks start once. And now you'll see why it uses so few tokens. The moment Claude clicks start, it stops. Its job is done. Last time, Claude made every clip and did every edit itself. This time, it just sets everything up and hands it over. Now the tool takes over. It makes the clips with the same stick man in every scene. It puts each narration line at the start of its scene, cuts every clip to the exact length of its line, and joins everything into one video. And when it finishes, look for this: alignment okay. Every narration line starts inside its own scene. Here's the result: story, character, scenes, narration, and timing. All from one command, one steady voice, and no watermark. Then I click download, and the full video saves as one MP4. And when it's downloaded, I tell Claude, "Sign me out of Google." And Claude signs me out, right in the browser. Only do this after the download. If you sign out in the middle, the tool loses its login, just like what happened in my run. This one is about a minute because I wanted to show you how it works. For a longer video, just ask Claude for more scenes. More scenes use more flow credits, so check your balance first. And here are a few topics you can steal. How did ancient people find water? Why did people start farming? How did the first boats cross the ocean? Just change the topic and one word in the voice names, like Karen water or Karen farm. Remember the repeating voice problem? This is how you stop it from coming back. A quick note before you start. It's not instant. This one took about 25 minutes, so start it and go do something else. And don't run it back-to-back. Give it a break between videos. Now, as promised, let's build the tool. Inside Google Flow, open tools and click create tool. This is Flow's tool builder. You describe what you want and Flow builds it. I spent hours testing this tool and fixing every problem I hit. The repeating voice, the silent gaps, the wrong story showing up. Then I put every one of those fixes into one single prompt. So, all you do is paste it and send it. One paste. Flow builds the whole thing. The story box, the same stick man in every clip, a voice line for every scene, the duplicate check, and the timing engine that matches every clip to its voice. When it's done, click done and pin the tool. That pin matters because my command tells Claude to open your pinned tool. That's it. You build it once. After that, it's one command and a new video. Now, I want to hear from you. Comment below and tell me what you think about this tutorial, and I'll send you all the prompts you need. The tool prompt, the main command, and subscribe because next, I'm pushing this even further.
Could not fetch transcript (it might be disabled or unavailable). Error: Could not retrieve a transcript for the video https://www.youtube.com/watch?v=9Ky5QTpVjAw! This is most likely caused by: Subtitles are disabled for this video If you are sure that the described cause is not responsible for this error and that a transcript should be retrievable, please create an issue at https://github.com/jdepoix/youtube-transcript-api/issues. Please add which version of youtube_transcript_api you are using and provide the information needed to replicate the error. Also make sure that there are no open issues which already describe your problem!
Jacine, thanks so much for joining me today. Really excited to talk to you. And for those of you who don't know, Jacqueline is a product director at Google Labs, arguably one of the fastest moving teams inside Google. Is that is that fair to say or? >> We do value our ability to move fast and get things into users hands. >> Awesome. Awesome. So, let's talk about Gemini 4 is is a big launch that happened, you know, a week or so ago. And how has it kind of changed your workflow as a PM and as a team? >> Yeah, >> you know, I would I would say it's kind of changed it in two big fundamental ways. The first is just expanding the surface area of what's possible to build and what's possible to exist and what's worth existing these days. And I think that's taken a moment for us to really recalibrate as a product team and just think about what are the things that are worth building. So there's a lot of time we've been spending around just the ideation phase and trying to plan out what's next and figure out what can we now achieve in this space. And then the other is the actual day-to-day tactical parts of the how do we do our actual job. And on that front, a couple things that we've kind of shifted towards I've noticed. Number one, with a model like this and as the models get more powerful, it's really no longer a question of can the models do it or can we build it. It's a question of is it worth existing? Should we build it? >> And so instead of starting with what I often refer to as the 40page PRD, of which I have written many of those early in my days as a PM, what I'm often times finding is better to start with is more of a highle strategy doc. something that really asks the question on what are we building, why are we building it, why is it important, ask the bigger questions and gets everybody aligned on the opportunity space. In addition to that, I find it helpful to create a presentation cuz sometimes you want a document for offline reading. Other times you're actually presenting in front of a small group and that could be in an actual product review or it could just be like bumping into somebody in the MK or in the hallways. I know we had lunch earlier and like you just see the people that are walking around the office and sometimes I like pulling open a presentation and like bouncing ideas off of folks that way. And then the last one is a prototype before you actually want to go seriously in depth and build a production grade anything. Building a prototype is just the fastest way to gauge excitement. So those are kind of like the three things I'm finding are where the product journey starts these days. >> Awesome. And labs is like kind of the experimental group in Google, right? trying to prove our ideas and you have products for SMBs, you have products for consumers and all kinds of crazy products. Like how do you just how Yeah. Like you're totally right. It's not about building, it's about deciding what to build. So do you guys have some sort of methodology to decide what to build or is just kind of like, [laughter] you know, what's the most inspiring idea? Yeah, >> I think it's kind of, you know, both an art and a science. I would say we have a couple great internal documents that Josh has talked about many times as well. We have some labs guiding principles. We also just have this general philosophy that we want to build users first, >> Google second, labs third. And so we look out in the world and we say what is worth existing at this point? What do users want or what do we think that they're going to want? >> And then we figure out is this something that makes sense for Google to be building and then is it something that labs should be actively contributing to? But I think when it comes to new ideas, there's usually like two lenses I look at them at. One is what is this emerging model capability that we're starting to see and does that open up a new type of product that we should be building as a result of that or does it cause us to revisit old ideas that didn't make sense before that are now viable and then the other is what are the new user trends we're sensing in the world. I think as we're all adapting to how AI is changing our day-to-day life, it actually causes new habits for people to create and that can open up new surface area or new white space for products or solutions to exist. And so like good oldfashioned userdriven product development, but with the lens of not trying to see what people are necessarily all doing today and more about predicting where you think things are going and taking a few bets on that line. Maybe we could talk about like a specific project maybe like Pomet or whatever project you want to talk about and talk about you know how did you decide to build for this group of people and what kind of problems you were solving for them. >> Yeah, I think that one was definitely more it was a combination of both of things coming together. We saw that these image generation models were just getting a lot better at being able to create onbrand content. I remember seeing even just like on my ex feed all of these crazy elaborate prompts to get image models to do things, but it resulted in something super cool. But you also know that like your average SMB owner, of which my parents both ran their own small businesses while I was growing up, they don't have the time to be constantly tracking everything that's happening and figuring out how to use AI and prompt the models in these specific ways and, you know, every time starting from a blank slate. And so in that case, it was definitely about, hey, this model's capable of doing these things. you know, >> we know there's a group of users that could really benefit from that and we think that there's a different product shape that we can form around it. And then you start to go a little bit deeper and you realize, hey, to generalize this capability and apply it to any user is very different than just like a one-off prompt that somebody spent a lot of time curating. And so then that gets you into the whole idea of, well, what if we really understood your business so that when we did want to generate things for you, it was a lot easier for the user and they could kind of be involved as much or as little as they wanted into those workflows. Yeah, that was kind of the genesis of a lot of thinking on that one. >> And yeah, like I feel like the number one mistake that I think Bay Area PMs make is just like wearing our ivory towers here. It's very nice here. And then you know, we just think everyone's going to use these terminal interfaces and like these like complex agent interfaces to prompt like a huge prompt. And I think what I really love about Pomei is like how it makes it really accessible to people who are less techsavvy and people who actually have other things going on. Right. >> Yes. I'm I'm curious. How much do you go off and talk to a bunch of SSMBs and local restaurants or how did you how did you decide like you know how did you kind of put yourself into their shoes? >> Yeah. No, we were very big on co-creation. >> Okay. >> In labs and in this case we talked to both people that worked in the marketing sector as well as businesses. Absolutely. We work with a small group of them. If you look back on all the launches we've had we almost always feature a brand and that's somebody that we've worked closely with. But we also have groups of trusted testers and before we launch anything we do a certain amount of user research and bring in actually one of my favorite stories is we'll occasionally have on-sites with the team and this team happens to be split between Mountain View and New York and the team's actually going to New York next week to do a like group onsite there and part of that workshop is going to be bringing in a few small businesses and actually sitting with them and using the tools and showing them the stuff that we're building and getting feedback directly from them because we think it's super important that it's not just our opinion that's impacting the product. Like we want the users who are using it. >> Yeah. >> Helping us co-create and figure out what to build and what's working and doing that iterative loop to make sure that what we're throwing out there isn't just a reflection of what's technically possible, but is presented in a way that's actually accessible to the users that we're going after. >> Yeah, I really believe that's probably my number one calling as a PM just to the co-creation thing. [laughter] and like you know getting them into the community and like just talking to them every day showing all the designs and stuff and just like you know just like it's kind of unfair you just get the best ideas from the actual customer like you know why why not >> I think it's interesting because that is so important and then the hard part now is also you hear everything they want >> but don't just give them exactly what they want like think about what the root problem is they're trying to solve and then solve it in a way that they probably wouldn't have thought to solve it and that's when you get that like delight factor the second somebody uses it you kind of walk away being like, "Oh, I didn't think that this was possible, >> but this is actually exactly what I wanted." Like, I think that's what we're always trying to strive for. >> That's right. That's why you need both. You need both your input and also like kind of your understanding of what the model's going and all that kind of stuff, right? >> Exactly. >> I'm curious with Gemini 4 and everything else. You talk about not writing 40page PRDs and thank God, but like is there still like a lot of like, you know, strategy documents or what whatever. Is there still like a lot of reviews and approvals and like you know let's check the boxes and everything or is it like you know a little bit more nimble now? I know Josh is very into like moving fast. >> Yeah, >> definitely into moving fast. I would say it's less about having to check the boxes, but I do want to be careful here because writing those strategy docs I think still does take a lot of time. >> In fact, I think the person writing it should sit with it for a while and make sure that what they're actually coming up with is worth other people's time to read. >> That's right. That's right. And so I think a lot of the work that happens is now just on that initial phase of like honing your thoughts and making sure that your strategy doc is not, you know, 10 pages cuz it's so easy to write 10 page documents these days. Like trim it down. Figure out what the narrative is. Figure out how to position it in a way that resonates with folks. And then once you get that going, like once you've really unlocked a good idea, it's easy to get people on board and build. >> This episode is brought to you by Whisper Flow. Whisper Flow saves me at least three hours a week and is one of my favorite AI apps by far. It's just [music] so much faster to dictate to AI using your voice than to type. You just talk naturally and it outputs clean, ready to send text. [music] Whisper Flow even removes filler words and formats your sentences for you. I use Whisper Flow for everything including drafting newsletter posts, writing product specs, replying on Slack, and more. It works on Mac, Windows, iPhone, and Android across all of your favorite apps. Try a free at whisperflow.com and use my code peterwisperflow to get six months free. That's peter whisperflow. Now back to our episode. Have you like come across we don't have to name any names but have you come across like these dogs where like you read halfway through and it's like oh this is written by AI like this like [laughter] do you have like a you know principle on that? Like you have to be able to maybe it's written by AI but you got to be able to defend the idea. >> Yeah I would say in the early days this was definitely more common. I think people are starting to adjust adjust and adapt, but I have a few rules at this point. And one of these actually I got from talking to somebody during an SF tech event last week, but I just love the way they summarized it, so I'm going to recote them, which was basically if it takes somebody longer to read the thing you produced than it took you to produce it, >> not allowed. >> Like a I think that just shows you didn't even take the time to read it yourself. And so that's like a big red flag in my books. But then the other thing I would say is you should be able to actually defend every single word that you've written out. And you should be able to have an active conversation about the idea. And if somebody reads a doc and they get frustrated because it feels like it's too lengthy or not like compact enough, I would push it back on them and say, "Hey, can you take another try at writing this?" And I think it just it takes a bit of a muscle to learn how to do that and to not rely on AI, but like to use it as a thought partner, but you really want your thoughts coming out in what you produce at the end of the day. >> Yeah. Like you should be able to you should get the message across in like, you know, like one page basically, right? You don't have to write like six pages or anything like you know. >> Yeah. I keep calling it a one pager. I feel like mine average one and a half these days, but like if it's getting beyond that, then you probably haven't honed the idea enough to effectively communicate it. >> Got it. Okay. All right. All right. Well, let me ask you maybe some harder questions. Let me ask you for labs. There's a lot of experimentation going on. One argument you can make is like, you know, it's great there's also experimentation, but why does the team just focus on making the Gemini app better or something like that? Or maybe let the experimentation actually lead the way to make the Gemini app better. >> You know, I think one of the great parts of Google is that we have a lot of amazing teams that are working on different products. And so we have an awesome Gemini app team that is working on that product. And then we have a bunch of people in labs whose goal is to try to >> invent and build for the future. And sometimes it is actually a great story that's something that we've been working on helps make the Gemini app better. And that could be the actual product or experiment that makes its way into it. It could be just the learnings that we've had that help influence how the rest of us are thinking. So you see a lot of stories like that where something from labs might help influence the Gemini app, but also just the way we're trying to, you know, carve out the space in the future, I think generally helps overall with just the mindset and the culture and making sure we're all thinking with that that point of view. >> Yeah, I think notebook LM is now part of the Gemini app, right? >> Yeah, you have Gemini notebooks in the Gemini app. You have CC was an experiment we launched last year initially and that kind of helped turn into the daily briefing that's in the Gemini app. And then that's actually a great story because I recently launched a family CC as well. So that kind of shows >> there's many paths to success for what a lab's experiment is and I could probably list out like five or seven other ones at this point too. >> How do you guys decide whether a lab experiment, you know, should be wound down or like mix it into Gemin case by case? >> I think it always ends up being case by case what makes the most sense. But more often than not, it's the team itself that's building it that happens to know what the right path is for that. And that comes down to we want the people working on these so deeply embedded in the problem space that they're trying to solve. And it's it's them that kind of help us figure out, all right, like where is this going? Does it grow? Does it graduate? Does it, you know, reh harvest into the next great thing? But a lot of just autonomy and relying on the fact that these teams know what's best for the products and experiments they're working on. >> I see. Interesting. But like I feel like as a PM it's kind of hard to kill your own product, right? Or like you know like I don't want to give up on it. Like do you guys have some lab principles or like to kind of encourage you know objective thinking around this stuff? >> Yeah, I mean I think it's definitely a muscle that grows with you >> the more you're in labs and that is part of the culture that we really try to curate here. We have a few fun sayings that are helpful to point back to. I think I've said one already like build to learn. We just very much believe in building to learn. We also another favorite saying of mine is keep the lessons not the prototype. >> Like you learn so much as you're building things. And I think a key thing is never look at an experiment ending as a failure. >> Like you always take something away from it and it's helpful for the teams to do a reflective piece on like what are all the lessons learned? How can we share that out with the rest of labs? So we've also done things like we have our labs live which is our whole team meeting that happens periodically and during some of these we'll invite the experiments up that have since moved on to come and talk about all of the things that they learned and also invite other people in labs who might now be exploring a similar area but with a different lens or a different like point of view on how it might be worth solving it to talk to the team like how do we continue to pass on the learnings as well as sometimes it's just the infrastructure and like the actual code that was created. is still able to be reused in other areas. But I think just celebrating learning is a really key part of it. >> Yeah, I think that's like really important like the culture and the incentives like because people want to build their careers and stuff. You got to make sure that >> like I always believe that if you want to build a career idea you you don't have to think about all the career performance stuff. You just focus on the product and make it good >> or or like you know learn something and then you build a career naturally, right? >> I very much agree with that. >> Yeah. But I think as big companies it's kind of hard to like just refocus on that. you kind of have to encourage that, you know, behavior. >> Yeah, you definitely have to encourage it. I think, you know, a lot of credit to our VP Josh who has just done such a great job helping to create this culture and this, you know, natural curiosity. And then what I found is like you tend to attract the types of people that do actually thrive >> in these areas. And so like the more culture and the more you prioritize building for that team like the more you'll be able to continue to like recruit and the right people tend to find you which then helps like in a positive way to continue on those traditions. >> Yeah. He kind kind of sets the example to a certain extent, right? I think he started here as an intern or something, right? >> I think so. Yeah, he's been here a while. [laughter] >> Yeah. Yeah. Let me ask you a question about Josh. Like he seems like such a nice guy. What happens when he actually gets a little bit disappointed in you? like does he like get mad or what? Give them some feedback or >> he's just such a positive person. I think >> what he does is he tries to empower the team. >> And I mean like I think if you love the products that you're building like you're going to be the ones that hold yourself to the bar that you want to ship at and that's kind of the culture we've tried to create. So I've >> Yeah. like it's I don't know how else to describe it except that I sometimes find like I'm my own harshest critic >> and that as a result means that like I'm always going to try to do things and if things don't go or if things take a different turn you learn something and you're like oh this thing I thought was going to work out actually there's a whole bunch of challenges it's a safe place to talk about it and then like we're all we're all in this together so like how do we course correct like how do we move on to something different I think there's always a positive way to look at situations because at the end of the day we're all here to try to build great things for the users. >> Yeah. And then it's not like you know it's not like life or death or anything. It's like we're trying to you know >> we're building Yeah. We're building great products, great experiences and everybody is motivated to make that happen. So >> if anybody ever like similarly is >> has a question or like feels stuck on a team, there's always people in labs that are there to help and you know share what they've learned about a space or like provide input on like how they might think about it. And actually it was great earlier today. We do a lot of these also internal uh meetings within the team to try and just knowledge share cuz there's so many of us building so many interesting things. So we call one of them M&M's which are I think methods and metrics and you'll have a team that will come and present something that they've done and it's a great way to just knowledge share out and a great chance for them to also let other people see what it is they're building. So that's one example of you should never feel like you're stuck or you don't know how to solve a problem. Like I always say go to AI to help brainstorm in that but also go to the people that you're surrounded with and that's something that we really tried to cultivate in labs. >> Yeah, that's what I actually really miss after doing my job. Like it's it's fine brainstorming with different models, but it's not the same as brainstorming with actual smart humans. It's just not the same, you know? >> I know. Well, that's why that's why I wanted to hang out with you today also, and that's why I try to make it a point to not just talk with people at Google, but also just like continuously go to different industry events and just like bounce ideas. I think there's a lot to be said about that. >> Have you with all this like I'm I'm assuming everyone's like, you know, using anti-gravity or using AI like all the time. I know in the engineering side like the problem becomes the bottl net becomes like people are almost pushing too much code and then like you know you get AI to do all the code reviews and then you don't even know what the hell's going on anymore if you're not careful like I'm curious on the prod side is there something similar like that is there like a lot of has a bottleneck shifted or >> I think >> what it's really shown is we need to think about what types of products to build >> okay >> differently >> you can just build 100 different things >> yeah like it's so like you can just build it doesn't I mean you should be building but it's also what is the value of a product I think is shifting especially as we're moving towards this agent first world where now you have users that aren't just people but they're agents and is it the agent that belongs in the product is it a consumer agent that's using a product on behalf of a user all of that is in flux at the moment and it really just makes you want to realize that you should be building as we say like you want to build products not features and I think a lot of things that used to be products before could be easily reproducible throw a couple agents at it over a weekend. So, that's not a durable product. >> What is that's the question that we're, you know, bringing up more and more these days. >> Do you have any do you have any initial thoughts around that or >> I mean, I think there's some good frameworks to look at. You know, network effects and, you know, context is interesting to think about. So, I think there's like a few common ways to slice the problem. But I also think just a lot of that is changing in real time. And that's why I mentioned around like what are new user patterns that are emerging and like if you play that out and assume a future that you know executes on that like what new problem spaces open up and then you can figure out products that might fit in there but also figure out what is the durability of them and the code itself is not the durability of them. Yeah, code is going to be like kind of just abstraction, right? Like I think I don't know. I feel like as a I've I've also been a PM for a long time and I've always prided myself in like you know caring about the UI and like making the user experience great and kind of stuff and now we're moving to a world where I think San Francisco and the Bay Are these people are just using their agents like they're even looking at UI anymore. They're not they're not going to your website or download your app. They're just like using some API or something. I think it really makes you question why do you want a person >> in your product? Like is your product something that needs a UI? And I do think in some cases it absolutely is. I think there's a lot of things that people will want to still be involved in. Anything that involves me creating something, I have a lot of fun doing creativity projects with my kids. Anything where I want to learn something, there might be a reason for me to interact. Anything I want to or anytime I want to be entertained or consume something, I'm still the one using the product. But you're right, there are definitely products that I use today that I would be more than happy to not use or at least not be the main person interacting with the interface, but perhaps the underlying capability or what that product or service offers is something I would still want. >> But that's what I mean where I'm like there's a lot to be thinking about differently when it comes to building products. >> There's definitely like a ton of like just like manual like just filling out an online form or like do some PDF or tax document or something that's that's like just stuff that we don't want to do, right? So I think agents can help a lot of that and especially like SMBs that you were building for Pomei, they have a lot of paperwork. >> Yeah. >> All the toils and tasks as we like to say. >> Yeah. So I definitely use AI to handle all that stuff. But then at the same time sometimes when I like give feedback to someone like I won't mention what the lab is but like the lab and then their agent respond to me instead of the human and then I get a little bit offended. I'm like hey why did you respond to me yourself? Why do you get your agent to respond to [laughter] me? So it's kind of like a double-edged sword cuz I'm using the stuff myself too you know. I mean, I think that's a great example of user behavior is shifting >> right now. And I think we're right in the thick of it at the moment. And it's interesting to think about how this all plays out and what are the products and experiences that should exist once we've all kind of gotten used to this whole new set of capabilities and how we want to integrate them into our lives. >> That's right. Yeah. >> Yeah. I'll give you another like quick silly example. I I were chatting earlier. So, I have three kids and we love hiking in the mountains and I took my daughters up hiking near Yusede and I recorded it on Straa and my three and a half-year-olds, this is the first time they did this Yusede hike all on their own and it was like a pretty lengthy hike and at the end of it I was really frustrated because although it was all stored nicely in Strava, there was this like weird jut out on the GPS that was just like a glitch. Okay. >> And in order to fix it, I did look into it like you have to download the actual GPS like upload. You have to like splice the section that doesn't work. You have to like stitch it back together. You have to reupload. It's like this huge manual process. And like that's the perfect thing where it's toilome. I want to just throw my agent at it. And full disclosure, I absolutely did throw my agent at it. It fixed everything. >> But I do still like seeing a feed of what my friends have accomplished. >> So it's not that I don't want to use Strava. It's just I didn't want to use it in that particular case. And as I look at products and what you're building, there's questions around like what are the toilsome things that users might need to do? How do you make them not have to do that? And then how do you play into more of what they might want to be spending their time doing in your product? >> Yeah, we were just talking about during lunch about how I think both our daughters like to paint and draw. >> Yeah. >> And like you can easily imagine a future like even like a year from now where you mostly focus on her painting and you just fire some commands to the agent to do your work. [laughter] You go back to painting or like you know DD these days when I work a lot of times I'm like actually outside just talking to the agent and I I think it's like a big quality of life improvement to just like stare on my screen the the whole time. I am such a big fan of voice. Like that has become my primary way of getting thoughts out of my head and into whatever I happen to be using. Like you should see my voice memos at this point. It is just and I love going for walks too cuz you're right, it is a quality of life improvement where I can literally like do laps around the building and just like talk to offload my thoughts and then I get back and I'm like, "Okay, now I can massage this into exactly what I want to turn it into." Most my strategy docs start off this way actually. >> Yeah. Cuz your mind's clear when you're outside and just ram blank, right? just yapping. This is >> scary. >> 100% 100%. >> But then maybe our neighbors are like, "Who heard these weird people just like yapping to their like talking to themselves?" >> You know, I've had that thought more than once. But it's okay cuz I feel like people have always talked into their, you know, headphones when they're on phone calls. >> So now I just assume they think I'm talking to somebody [laughter] on the other side right now. >> That's so funny. That's funny. Yeah. Well, I guess now let's move to like the family side of the interview. So, you know, that's pretty like techsavvy, I guess. Parents, like how are you thinking about raising your girls, you know? Yeah, that's a great question. I know it's it's on a lot of parents' minds these days. >> There's a couple things I'm trying to be really deliberate with. >> Okay, >> first, as much as I love AI and lean into it, and we'll talk a bunch about that, I do think it's still incredibly important that my kids get outside and that they learn how to be bored. I talk a lot about trying to also manufacture friction for them because I think it just teaches you skills that are necessary in life and that will be helpful in life. And in a world where things are easier and easier to do, I want to make sure that my kids still have that muscle to endure friction because there will still be frictional points that occur in your life. >> But then when it comes to how I incorporate AI into my day-to-day with them, for context, my eldest just turned five and my twins are three and a half. I want them to grow up in a world where they think anything is possible. >> Like I want to use AI to make their wildest imaginations come to life. Mh. >> So, a couple side projects I started with my eldest. We started a scary stories podcast because she loves scary stories and we started running out of interesting podcasts to listen to on this. So, I was like, well, why not try to make one with her? So, she helped me literally choose out all of the voices. We had some fun doing like voice design. I was using 11 Labs for this one. Although the new Gemini TTS model, I've been using that also. And like voice design is fun to play around with. But she chooses all the voices. She names the characters. I then use an image genen model to generate different images of the characters and like she gets to help me refine them. She gets to help me come up with the story lines for each episode. And then once we get the first version of the podcast done, she listens to it and she tells me what she likes and if there's any changes she wants to have made and so then I'll send it off to my agent. It'll remix everything and once her and I are happy with it, it's, you know, now posted and live on Spotify for others to listen to. >> Spotify. >> Oh, you want to know why I did that? because initially I was starting to make music with her using flow music and we made this really funny song about banana slugs after going camping in Redwood National Park >> and one of the first things she asked me was it real? >> I was like what do you mean it is it real? Like we're sitting here in the car listening to a song we just made like yes it's real. She's like no but like can my friends listen to it? >> And I was like to her real is a shared experience with her friends. Spotify and she'll grow up in a world where stuff's also sharable and shared with people. >> That's super cool. Yeah. >> Yeah. So, that's been one fun one. And then I've been having a bunch of fun with Argon over the past few days. Also, we built my kids love marble runs, the videos of them. And my husband likes it cuz intuitively he's like they're learning physics and he's really big into physics. And so I had Argon literally oneshot a marble run video for them that they got to see. And next I'm going to like customize it with them so they can choose the colors they want and feel like they're shaping it. And then this morning I because I'm a mother of three young kids, I wake up randomly early. And so as I was getting ready to come to work today, I was like, "Oh, what would be really cool is if I could build an interactive like petting zoo game for my kids." And so in a couple prompts, I ended up building this like whole interactive kind of game that you can walk around. It has six different animals and they're very like esoteric animals that the model decided to put in this petting zoo, but you can walk around and as you approach them, it tells you a fun interesting fact about them. >> Oh, that's really cool. Yeah. >> And so then that gets you into, well, how are my kids going to learn as they get older? And like, can we create these bespoke experiences for them? Yeah. >> And it's just really fun to see them grow up like with this being a normal thing for them. >> Yeah. Kids are kind of spoiled these days, right? Cuz like when we have to learn, we have to read like dry texts. But now you can just make like, you know, clips and videos and games of everything. You can learn everything through that. >> I mean, any way that gets my kids excited about learning, I am more than more than happy to indulge that side of them. >> Yeah. I think all your stories share a commonality and like I think it's important to try to get the kids to go to creation as opposed to consumption. >> Yes. You know, like like when my kids watch like these like shorts, I'm like it's like very brainwashing. But like when they actually make their own games and actually you don't actually need to type to make games, right? You just like speak to a phone or speak to the computer and they'll make a game for you. >> And when you do that kind of stuff, I'm like if you make your game and you make it good, you can have like unlimited screen time in like making it good. You know, you can just do whatever you want. >> Honestly, I think it just connects different parts of your brain when you're actively involved in the creation point. And like 100% agree with like if you're being productive and you're learning and you're challenging yourself, like that's what I want to see them using AI for, not just like as a way to make their life easier right now. There will be plenty of time to smooth those things out later. For now, I'm like I want you actively learning and being curious about it. And then also expanding your mind cuz if you're growing up thinking that all of this is just normal and natural, like you said, we grew up in a very different era. like I can't wait to see what the next generation comes up with and like what they bring into the world. >> Mhm. I also think like it's very important to yeah to pursue your curiosity and also show that you can turn your curiosity into like a real career you know like when I was younger like I I make games on like Apple 2 or whatever like the old Apple games [snorts] but I always thought like it was like not a real career. I was like okay you know I have to go get a real job. I have to go become like a investment banker or something. But I actually want my kids to like not only explore their creativity but also like you know we're talking about like actually make some money online and see how it goes so that they can actually be ambitious and pursue whatever they want to do you know. >> Yeah. I think it's important for them to connect the dots on right now your kids are a little older than mine. So I'm like curiosity is the main thing that I'm pushing them to think into. But absolutely how does that tie into the real world and like where do they want to spend their time? What do they want? And how do they help create the life that they want to live at the end of the day? um is it's important lessons. >> Let me wrap up with this. Now that we have, you know, Gemini Argon and I'm sure have a bunch of like Gemini 5, Gemini 6 internal models going on, what are you most excited about, you know, building labs or with your kids or what are you really excited about? Yeah. >> Okay, I'll start with the kids side because that's easy and then I'll answer the lab side. With my kids, I'm just like I am a neverending faucet of ideas of things I want to be building with them. And I'm really excited as they get a little bit older for them to be even more active participants in the ideas that we're making. And I just like there's so many things I want to run home and just start building with them right now. Time is like my limiting factor these days. >> And then in labs, I just this next class of models that feels different. The possibilities are so much bigger. And I'm really excited that we get to be in a place where we can think about what that future is and start working towards it. >> Yeah. >> And I'm just so grateful that like this is my job and I have an awesome team that's also thinking about these really interesting hard problems and coming up with like some really creative solutions. like the best solutions and the best ideas I find somehow connect the technical capability with that user need but come at it from an angle that you wouldn't expect and kind of make you go huh that's interesting >> interesting >> I kind of like it and I'm seeing more of that coming out these days. So yeah, just generally I think it feels different right now and I'm excited for all the new stuff that we have cooking. >> Okay, we're going to keep it competition until it's actually >> Yes. [laughter] >> Okay, got it. Yeah, I think I think almost like any kind of persona you have like any kind of customer you're building for I almost feel like these models at a point where they can almost kind of you hire a person, you just like solve your entire problem for for you, right? It's not like it's not like doing a task anymore. It's just like you just hire a person to just take care of this, you know? I just that's the way I'm treating a these days. I'm like instead of like doing a customer support task, I'm hiring a customer support agent just take care of everything. Yeah, I think it's it's an interesting behavior shift even in the way you're talking about it right now versus what likely were talking about AI a year ago, even myself. And you're right, I think like us >> as users are we're changing our expectations. They're rapidly shifting and the way we're thinking about how we want to apply it into our lives is changing. I will say I think there's different users and they all have like a diversity of what it is they want but user behaviors are definitely shifting and that's been something to calibrate against constantly as we're thinking about those new ideas. >> Awesome. Yeah. And I think just to wrap up I feel like part of our job or like part of part of like maybe labs in my opinion is like just teaching users who are less techsavvy that they can actually ask a lot from these models. Like you can actually ask a lot and they will just do it for you. >> Yeah. And I think it's more than just asking a lot. It's like how do we show them? How do we give them the experiences that they want? Cuz when you ask, that's exactly where people get stuck is they don't know. Yes, the blank prompt box problem basically. How do you solve that in a delightful way so that people are getting the most out of them cuz there's so much value that can be gotten out of them. >> Awesome. Awesome. Well, where can people find you online or you know? >> Yeah, absolutely. You can find me on X at Jackalu. Don't ask. There's a backstory behind that [laughter] one or on LinkedIn as well and Substack for that matter. >> Awesome. Thanks so much for your time. Yeah, >> thank you. Blessed and pray.
Anthropic, the company behind Claude, just made a move that could change capitalism forever. They announced that they're putting 5% of the entire company into a new fund dedicated to the public good. And if the company keeps growing, that share goes up. When you add in a pledge from the founders who promised to give away 80% of their own stake, and a program for employees, Anthropic says roughly 20% of the whole company is now set aside for public causes. And this is a company that's reportedly heading towards an IPO at a valuation of around $2 trillion. That means the 5% alone could be worth something like $100 billion on paper. And then, the very next day, Elon Musk responded by saying that companies should give their shares directly to the public. And he pledged to give one of his SpaceX shares to every single kid in America, saying it will be the first of many. So, in this video, I'm going to break down exactly what was announced, why AI companies might be able to do something that no company in history has ever really done, what this means for jobs and for your future, and why I think this crazy-sounding idea is probably going to become a lot more normal. Starting with point number one, which is what Anthropic actually announced. So, let's start with the basics, because I want this to make sense even if you've never bought a stock in your life. When you own a share of a company, you own a tiny piece of that company. If the company becomes more valuable, your piece becomes more valuable. If the company makes money and shares it out, you get a slice of that money. That's really the whole idea of capitalism in one sentence. The people who own the companies get the rewards when those companies succeed. [music] Now, normally, a company's shares are owned by its founders, its employees, and its investors. And when a company becomes hugely successful, those are the people who become incredibly rich. What Anthropic has done is take a big chunk of the company and say, "This part is not going to belong to any of us. It's going to be used for the public good." Specifically, Anthropic announced something it calls the Anthropic Public Benefit Fund. They're committing 5% of the company's shares to it, subject to approval from their shareholders. But, here's the really interesting part. They also said that for every 50% increase in the company's value above a certain starting point, the board can add another 2% of the company to the fund. In Anthropic's own words, "The more Anthropic grows, the larger the share of the company that can be dedicated expressly to the public good." So, the more successful they become, the more of the company gets given away. That's basically the opposite of how most companies work, where the more successful you become, the more the people at the top keep for themselves. Here's a simple way to picture it. Imagine a bakery that's owned by its founders and a few investors. [music] Every slice of the bakery belongs to one of them. Now, imagine the bakery announces that one slice of the business now belongs to the town. Whatever that slice earns and whatever it becomes worth, goes towards helping people in the town. And on top of that, every time the bakery grows by half, it hands the town another slice. If the bakery stays small, the town gets a little. But, if the bakery becomes the biggest bakery in the world, the town ends up owning a huge piece of something incredibly valuable. That's basically what Anthropic is promising, except the bakery is an AI company that might end up being worth trillions of dollars. They've also said a few other things about how this fund will work. It's set up as its own company, rather than a traditional charity, which Anthropic says gives it more flexibility in how it gives money out. It could make huge donations to existing charities, but it could also do what they describe as direct distributions to specific groups of people. In other words, it could potentially send money directly to people who need it. And they made a point of saying that none of this money can be used for lobbying or political donations anywhere in the world. And they gave one really important example of what the fund might focus on, and that's the impact of AI on jobs. Anthropic has already committed $200 million to this issue, and they said the fund could pursue bolder approaches. They wrote that AI may generate enormous economic value, but it could also cause instability for many workers. And I think that sentence is really the key to this whole story. Because when an AI company starts setting aside a fifth of itself and specifically mentioning workers who might lose out, that tells you something about what they think is coming. Which leads me to point number two, which is the founders pledging 80% of their own equity. Because the 5% is the company's pledge, but the founders made their own personal pledge, too, and it's even more extreme. According to Anthropic, when the company was founded, its co-founders pledged to donate 80% of their equity. That's all seven of them, including the CEO, Dario Amodei, and the president, Daniela Amodei. And reports say this pledge is now written into the documents the company has prepared for its IPO, which is the moment a company starts selling its shares to the public on the stock market. Now, let's put that into perspective. If Anthropic really does go public at a valuation anywhere near $2 trillion, the founders' stakes could be worth tens of billions of dollars each. And they've pledged to give away 80% of that. That's not a small donation to a local charity. That would make them some of the biggest givers in human history. For comparison, a lot of famous billionaires have signed pledges to give away at least half of their wealth over their lifetimes. These founders committed to 80% before the company was even worth much at all. And it doesn't stop with the founders. Anthropic also has a program where employees can pledge some of their own shares to charity, and the company matches it. Anthropic says about half of its employees take part. So, when you add all of this together, the company's 5%, the founders' 80%, and the employee program, you get to roughly 20% of the entire company being set aside for the public. And Anthropic said something really interesting about this. They said they're directing equity that tech companies often use for ever-increasing executive pay towards the public good instead. Now, of course, not everyone is convinced. David Sacks, the investor and former White House AI czar, posted on X that Anthropic had put 20% of its equity into an LLC, and that whether it serves the public good remains to be seen. And that's a fair challenge. We don't yet know exactly how the fund will be governed, how it will be taxed, or how much it will have to disclose. Anthropic says those details are coming in the next few months. So, it's right to be a little bit skeptical until we see how the money is actually spent. But, even with that skepticism, the scale of what's being promised here is genuinely new. Which brings me to point number three, which is Elon Musk's response. Because this is where the story got really interesting. Almost immediately after Anthropic's announcement, Elon Musk replied, and his message was simple. I recommend giving the shares directly to the public. In other words, why put the shares in a fund that someone else controls? Why not just give them to people? And then, the very next day, he went a step further. He posted on X, "Next year, I will give one of my SpaceX shares to every kid in America." And then he added, "This will be the first of many." Now, let's think about what that actually means. There are roughly 72 million children in the United States. At SpaceX's recent share price, which has been around $146, that works out to something like 10 and 1/2 billion dollars worth of stock given away by one person in one go. And he's saying, "This is just the beginning." And this isn't the first time someone at SpaceX has done something like this. Back in July, SpaceX's president, Gwynne Shotwell, and her husband pledged to donate one SpaceX share to each of roughly 2 million children who had already been signed up for one of these new children's savings accounts. So, there was already a small version of this idea happening. What Musk did was take it and scale it up to every child in the country. And if he really means it when he says this will be the first of many, then over the years a lot of American kids could end up owning a real piece of one of the most valuable companies in the world just by being born. Now, there are a lot of unanswered questions. Musk didn't explain exactly how the shares would be handed out, who would count as a kid, or how families would actually receive them. One idea that's been floated is using what are called Trump accounts, which are new savings accounts for children that the government recently allowed people to donate shares into. Brad Gerstner, the investor behind that idea, publicly praised the move. But for now, it's a stated intention rather than a finished plan. And this connects to something Musk has been talking about for a while. He's argued that instead of a universal basic income, the world will need what he calls a universal high income. The idea being that if AI and robots end up doing most of the work, the wealth they create needs to flow back to people, not just a small group of owners. Earlier this year, he wrote that universal high income via checks issued by the federal government is the best way to deal with unemployment caused by AI. So, when Musk talks about giving away shares, it's part of a bigger picture he's been painting about what happens when machines do most of the jobs. And whatever you think of Elon Musk or of Anthropic, I think it's genuinely remarkable that two of the most powerful players in AI and space are now publicly competing over how to give their companies away to the public. A few years ago, that would have sounded completely absurd. Which leads me to point number four, which is why AI companies might be able to do this when no company has before. Because I think the obvious question is, why would any company give away a huge chunk of itself? That's not normal. Most companies fight to keep every single percentage point of ownership. So, what's different about AI? The answer comes down to just how much money these companies might end up making. Let's think about it simply. Most businesses make money by selling a product or a service, and to grow, they usually need to hire more people. If a law firm wants to take on twice as many clients, it generally needs a lot more lawyers. But AI is different. If an AI system can do the work of a lawyer, an accountant, a programmer, or a customer service agent, then the company that owns that AI can grow enormously without hiring many more people. The work is being done by software running on computers 24 hours a day. Think about what that looks like in practice. In the past, if you wanted to build a company that served millions of customers, you needed thousands of employees, customer support teams, engineers, accountants, designers, sales people. In the future, it might be possible for a company with 10 people to do the work that used to take 10,000 because AI agents are doing most of the actual tasks. And when 10 people can do the work of 10,000, the money that used to be spread across 10,000 salaries ends up concentrated in the hands of whoever owns the company. That's great for the owners, but it's a huge problem for the other 9,990 people. That means the profits from a huge amount of work could flow to a very small number of companies, not just one industry, but potentially across the entire economy. Some people in the industry believe the leading AI companies could become the most valuable companies that have ever existed [music] by a long way. And if you're one of those companies, and you genuinely believe you're going to capture that much value, then giving away 5 or 20% might not actually hurt you very much. The remaining 80% could still be worth more than any company in history. And there's a deeper point here, too. In the economy we've had for the last couple of hundred years, most people get their share of the wealth through their jobs. You work, you get paid a wage, and that's how the money spreads out across society. But if AI starts doing a lot of that work, the wages start to shrink, and more and more of the money goes to whoever owns the AI. In simple terms, [music] the wealth shifts from people who work to people who own. And if you don't own anything, you could get left behind. So, that's why ideas like this are suddenly being taken seriously. If the future economy is going to reward ownership instead of labor, then one way to make sure everyone benefits is to make sure everyone owns a piece. That's essentially what Elon Musk is suggesting with the SpaceX shares, and it's one of the things Anthropic's fund could potentially do. And it's why I think, while this sounds like an insane idea right now, it's probably going to become much more common in the future. Which brings me to point number five, which is the fact that Anthropic is already using its own AI to automate itself. Because this is the part of the story that I think a lot of people are missing. Anthropic isn't just talking about AI changing other companies, it's already changing Anthropic. In a report from earlier this year, Anthropic said that more than 80% of the code merged into its products in May was attributed to Claude, its own AI model. And it said that figure is conservative, with company leaders estimating that 90% or more of their code is now written by Claude. Now, think about what that means. The people who work at Anthropic are some of the best programmers and researchers in the world, and the majority of the actual code is now being written by the AI, while the humans mostly direct it and review what it does. Anthropic said the typical engineer was merging about eight times as much code per day as they were back in 2024. They were careful to say that lines of code isn't a perfect measure of productivity, and that this probably overstates gain. But even so, it's a huge shift in a very short amount of time. And they've said that the quality of the AI's code went from being a bit worse than their human engineers to roughly equal, and they expect it to be better within the year. So, this is a company watching its own product become better than its own employees at one of the core parts of their jobs in real-time. And the same report went even further. It was called when AI builds itself, and it talked openly about AI being used to help build the next generation of AI. That's something people sometimes call recursive self-improvement, where each version of the AI helps create a smarter version, which then helps create an even smarter one. Anthropic said this is already starting to happen in small ways inside the company. And alongside that, it called for the world to develop a verifiable way to pause AI development globally if things start moving too fast. Now, think about how unusual that is. Here's a company saying, "Our AI is doing most of our coding. It's starting to help build its own successor, and we think the world needs an emergency brake." When you read the public benefit fund announcement with that in the background, it starts to make a lot more sense. These are people who seem to genuinely believe that a massive amount of change is coming, and fast. And that's exactly why the jobs part of this story matters so much. If the company building the AI is already using it to do most of its own work, it's not hard to imagine the same thing happening at banks, at law firms, at marketing agencies, at consulting firms, and at pretty much any business where people sit at computers. Anthropic itself has been pretty open about this. Dario Amodei has warned in the past that AI could wipe out a large share of entry-level white-collar jobs. And that's a big part of why they keep bringing up workers and labor markets in announcements like this one, which leads me to point number six, which is what all of this means for the future of jobs. So, let's talk about jobs, because I know that's the part that most people actually care about. And I want to be honest, nobody knows exactly how this is going to play out. But there are a few things that seem pretty clear. The first is that the jobs most at risk in the short term aren't factory jobs or manual labor. For a long time, people assumed automation would hit those first. But AI is actually best at the kind of work that happens on a computer. Writing, coding, analyzing, summarizing, answering questions, filling in forms, creating reports. That covers a huge amount of office work. And a lot of the entry-level versions of those jobs, the ones that young people normally use to get started in their careers, are exactly the kinds of tasks AI is getting really good at. The second thing is that the public is already nervous. A poll from late September found that around 64% of American adults think AI is developing too fast. Only 8% said too slow. So, people are clearly feeling that this is moving quicker than society can adapt to. And the third thing is that if this does happen, the old safety nets probably aren't designed for it. Unemployment benefits are built for people who lose a job and then find a new one a few months later. They're not built for a world where entire categories of work start disappearing all at the same time across the whole economy. That's why companies like Anthropic are suddenly talking about funds, distributions, and labor market programs. >> [music] >> And it's why people like Elon Musk are talking about universal high income. Now, I don't want to be all doom and gloom here, because there's another side to this. Every big technology in history has destroyed some jobs and created others. The internet wiped out a lot of jobs, but it also created millions of new ones that nobody could have predicted. AI could do the same. It could make people far more productive, create entirely new industries, and make a lot of things much cheaper. The big question is how fast all of this happens, and whether the benefits get shared, or whether they just pile up with a small number of companies. And that's exactly where the idea of an AI tax comes in. But before we get to that, I want to give you something practical, because I know a lot of people watching this are wondering what they should actually do about their own jobs. And I think the best advice right now is pretty simple. Learn to use these AI tools properly in whatever work you do. The people who are going to struggle the most are the ones who ignore AI completely and hope it goes away. The people who are going to do best are the ones who figure out how to use AI to do the work of three or four people, because those are exactly the people every company is going to want to keep. Focus on skills that involve judgement, trust, relationships, and taking responsibility for decisions, because those are the things that are hardest to hand over to a machine. And if you're young and just starting out, try not to rely on a career path that's built entirely around tasks a computer could do. That's not meant to scare anyone. It's just the reality of where things are heading. Which brings me to point number seven, which is the idea of an AI tax and other ways the wealth could be shared. So, if AI companies end up making a huge share of the world's money, how do you make sure that money reaches everyone else? Voluntary pledges, like Anthropic's fund or Elon Musk's shares, are one way. But a lot of people think that relying on billionaires and companies to be generous isn't enough. And that's where the idea of taxing AI comes in. There are already a few different versions of this idea floating around. Earlier this year, OpenAI published a policy paper suggesting that governments should shift taxes away from labor and towards capital. Which basically means taxing ownership and profits more and work less. It also suggested creating a public wealth fund where the returns could be shared with citizens. A bit like a national savings account that everyone gets a piece of. Then you've got much more aggressive ideas. Senator Bernie Sanders introduced a bill this year that would put a one-time 50% tax on the stock of large AI companies and put those shares into a public fund. That's a massive proposal and it's very unlikely to pass in its current form, but it shows how seriously some politicians are starting to take this. Dario Amodei himself has written that a universal basic income could be paid for with taxes on the companies that benefit or by raising taxes on capital gains. [music] So, even some of the people running these companies are saying, "Yes, we should probably be taxed more." Now, there are some real problems with an AI tax. The biggest one is that it's really hard to define. What counts as an AI company? Almost every company is going to use AI in some way. And if one country taxes AI heavily, companies might just move somewhere with lower taxes. There's also the risk of slowing down progress in a technology that could do a huge amount of good. So, it's not a simple fix. But, I do think some version of this is coming. Maybe it's a tax, maybe it's public wealth funds, maybe it's companies being expected to put a slice of their ownership into the public's hands. And honestly, it might be a mix of all three. >> [music] >> We already have real-world examples of this kind of thing. Norway has one of the largest sovereign wealth funds in the world, built from its oil money, and it's essentially a savings account for the entire country. In Alaska, residents get an annual payment from a fund that's built on the state's oil revenue. So, the idea of a natural resource that belongs to everyone and whose profits get shared isn't new. The question is whether intelligence becomes the new oil, which leads me to point number eight, which is how AI could change capitalism forever. So, let's zoom out and think about what all of this could mean in the long run. Capitalism, as we know it, has always relied on a basic deal. Companies need workers, workers get paid, and that money flows through the economy. Business owners get rich, but they also need everyone else, both as workers and as customers. AI could break that deal. If companies don't need as many workers, the link between work and income starts to weaken. And if that link breaks, then something else has to take its place, or a lot of people get left behind. What Anthropic and Elon Musk are doing could be early versions of that something else. Imagine a world where every child is born owning a small slice of the companies that run the AI economy, where every citizen gets a payment from a national fund that owns part of the biggest AI companies, where companies are expected to give away a bigger share of themselves the more successful they become. That would be a very different kind of capitalism. Not the end of capitalism, but a version where ownership is spread out a lot more widely. And think about what that could mean over a lifetime. If a child is given a share of a company at birth and that company keeps growing, by the time they turn 18, that share could be worth a lot more than it was when they got it. Now imagine they get a share from several different companies every few years for their whole childhood. That's a small growing pot of wealth that every single person starts life with, no matter who their parents are or where they're born. It wouldn't solve every problem and it depends completely on those companies doing well, but it's a very different starting point from the one most people have today where if your family doesn't own anything, you start with nothing. And I want to be clear, there's a lot that could go wrong here. Pledges can be broken, funds can be badly run, companies can use generosity as good publicity while still holding on to the vast majority of the power. And there's a real question about whether a handful of tech companies should be deciding how the world's wealth gets shared in the first place. Some people would say that should be the job of governments and voters, not CEOs. So I don't think anybody should take any of this at face value. We should watch what actually happens with the money, but here's my honest take. I think this is one of those moments that looks crazy at the time and then looks obvious in hindsight. Right now, a company giving away a fifth of itself sounds insane. A billionaire giving a share to every child in the country sounds like a publicity stunt. But if AI really does end up doing a huge amount of the work that humans do today, then some way of sharing the wealth is going to become absolutely necessary. And the companies that are closest to this technology seem to know it. That's why they're moving first. And I think we're going to see a lot more of this, more pledges, more funds, more debates about AI taxes, and probably some pretty heated political fights about who gets what. Because the amount of money at stake is enormous and the decisions that get made over the next few years could shape how the economy works for decades. Who knows where we're going to be in 10 years, but I don't think it's going to look like the economy we have today. So, what should you watch for next? First, keep an eye on Anthropic's IPO because that's when we'll find out what this company is actually worth and what that 5% fund really adds up to. Second, watch for the details of how the fund is governed and how it spends its money because that's what will tell us whether this is real or mostly a headline. Third, watch whether other AI companies follow. If OpenAI, Google, or xAI announce something similar, then this stops being one company's experiment and starts becoming a new expectation for the whole industry. And fourth, watch what governments do. If politicians start using announcements like this as a reason to push for AI taxes or public wealth funds, then we could see the rules of the economy start to change much faster than anyone expected. Any one of those things on its own would be a big deal. All four together would be one of the biggest economic shifts of our lifetimes. And personally, I think the most important thing is that regular people are part of this conversation. Decisions about who owns the future of AI shouldn't only be made in boardrooms in San Francisco or in a handful of government offices. They affect everyone's jobs, everyone's savings, and everyone's kids. So, the more people understand what's happening, the better the chance that whatever system we end up with is actually fair. That's honestly one of the main reasons I make these videos because these stories might sound like business news, but they're really about the kind of world all of us are going to be living in. So, now I want to hear from you. Do you think Anthropic's fund is a genuine attempt to share the benefits of AI or is it mostly good PR? Would you rather have shares given directly to people like Elon Musk is suggesting or a tax and a public fund run by the government? And do you think AI is going to change capitalism forever? Let me know down in the comments because I genuinely want to know what you think. And if you found this video useful, make sure to like and subscribe because things in AI are moving faster than ever and I'll be here to break all of it down for you. Thanks for watching and I'll see you in the next one.
So, what's MCP and why does every AI tool seem to support it now? Your AI was trained on tons of data, but on its own, it's pretty isolated. It can't see your database, your files, or your GitHub unless something connects it. Before MCP, every AI app needed its own custom integration for every tool. So, three AI apps and four tools meant 12 different integrations, each with its own setup. MCP stands for Model Context Protocol, and it's an open standard for connecting AI apps to outside tools and data. It came from Anthropic, but ChatGPT, VS Code, and Cursor all support it now. I think of it like um a USB hub. Your AI is the laptop, and services like Gmail or GitHub just plug into it. There's three pieces. The host is the AI app you're using, like Claude Code. For each server, it creates something called a client to hold that connection. And the server is a small program that sits in front of a service like a database and tells the AI what it can do. A server mostly offers tools. A tool is basically a function with a name, a description, and the inputs it expects. And the host passes that list to the model. With the Neon MCP, I can just say, "Show me the collections in my database." The model picks a tool that runs a query. Now, here's the important part. The model doesn't run anything itself. The host might ask you to approve it, then the client sends the call to the server. The server runs a regular SQL query, sends back the rows, and the model answers with the real data. And um most of the time, you're not building a server yourself. A lot of the popular services already have one, and it's usually just a command or a few lines of JSON to connect it. So, at its core, MCP is a standard way to plug your AI into your tools and data so it can actually do things with them. Now, I'm working on a full MCP crash course, and I also go over MCP pretty thoroughly in my coding with AI course. So, hopefully this helped you guys out.
Could not fetch transcript (it might be disabled or unavailable). Error: Could not retrieve a transcript for the video https://www.youtube.com/watch?v=vhqq4U3mnNQ! This is most likely caused by: Subtitles are disabled for this video If you are sure that the described cause is not responsible for this error and that a transcript should be retrievable, please create an issue at https://github.com/jdepoix/youtube-transcript-api/issues. Please add which version of youtube_transcript_api you are using and provide the information needed to replicate the error. Also make sure that there are no open issues which already describe your problem!