Meta’s Muse Revival, Frontier AI Under Threat, The Rise Of Dopamine Sites

0:00 Did Meta pull off AI's biggest comeback now that Muse is a top the app store? And what does Muse' success say about frontier AI models? Do companies actually need them to win? Finally, we'll talk about dopamine sites. What are they and should we be using them? That's coming up on the Big Technology Podcast Friday edition right after this. Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional coolheaded and nuanced format. We have a great show for you today. We're going to go deep into the success or is it a success of Meta's Muse and talk about whether this is something that will end up being the biggest story in AI this year and whether Meta is actually back. We're also going to talk about whether Meta's ability to build Muse on non-frontier models is a threat to the frontier. And finally, we're going to talk about dopamine sites, which are sites where you basically shop for things, but it's just about the experience. Once you hit checkout and the items don't come and it's just for that dopamine hit and we're going to talk about whether we should be using these sites or not.

0:59 Joining us as always on Friday to do it is Ranjan Roy of Margins. Ranjan, great to see you. >> I am so excited that we're not talking about human extinction this week and we're talking about the latest in AI products too and not even model. We'll talk about models a little bit I believe at the end but we're talking about it's all the product is here. The product is back. And are you muse pill, Alex? >> I was about to say, you know, we've been big boys, we've been clawed heads, we've been chat chaps, maybe. I don't know.

1:29 But are we are we are we muse men? Is that what we are? Are we or I don't know what's mammates? >> No. >> No. It's that's letting them to define the the persona too much. I I think it's got to be either musies or muse men. But I'm willing to hear the argument for muse mates. Muses are I like musies. Whatever it is, >> musies. >> Are we musies? >> Okay, let's definitely get into it. I'm a instinct. I don't even know what word starts with I that would fill that as well, but muse and instinct. I the last few weeks have blown my mind and I'm excited to talk about it. Yeah. Okay.

2:08 Okay, so let's talk about so so you know as the world has been distracted by AI you know basically threatening to extinct humanity there's been a rise of these personal agents and they're actually starting to work well and instinct is one of them where you can basically you know use AI to go out and do things for you and you communicate it with it via your messaging app and then of course you know over the past couple weeks since the beginning of the month meta has released this this app called Muse which goes out and takes action for you and it's a consumer version of a lot of of these enterprise tools that we've been seeing and it's doing pretty well. this is from the Wall Street Journal. Meta shows AI agents aren't just for business anymore.

2:45 Early signs of success for Meta's new consumer focused AI agent are helping prove that AI agents aren't just the hottest thing in business technology. They may finally be gaining steam among consumers too. Just 2 weeks after Meta released the agent called Muse, the app to access it reached number one in the app store. And then over the weekend, X was filled with people with post by people praising the new agent. Shares in Meta stock even jumped 11% on Monday, the biggest one-day gain since April 2025 and its highest close since October. So the success of Muse were basically like you can go out and you can use it and ask it to do things for you and it you know takes another step than your traditional you know claude claude code or chat GPT codeex.

3:31 I'll just give one example and this is kind of how I became a musey this week. You know, my wife was looking for to book a specialist appointment in New York and she was doing the traditional thing of like log on to the website and go through all the phone numbers, you know, make phone calls and try to find out you know when the next appointment is and in New York it's just impossible to get a specialist appointment. So the only appointments she were finding she was finding were months out. So I was like all right this is a job for Muse. I told Muse what our healthc care is and to go find an appointment for this particular specialist in New York nights and weekends, you know, preferred so she didn't have to take time off for work and Muse went out and came back with a list of providers. I asked for their phone numbers into Google Maps and it came back with a list of their phone numbers in Google Maps. I copied and pasted from the app, dropped it into my wife's WhatsApp and the next day she had an appointment with the first specialist on the list you know that Muse recommended. So, it's able to go through and take this arduous work and this sort of the most annoying parts of the internet and actually get stuff done for you. I also had to like unsubscribe this person who'd been getting Bank of America mail for like years in my apartment and I just didn't have the time to like go to Bank of America and ask them to unsubscribe. And I said, "All right, I'm going to just take a photo of the letter hand it to Muse and see what Muse can do." And Muse was able to figure it out. Now it's it's basically, you know, effectively a refined version of OpenClaw, which was all the rage, you know, just a few months ago, but it seems like Meta has latched on to something here. Your thoughts, Ron, >> okay, I I have a lot of thoughts on this one. I think I wanted to start with as you said and even as this article is alluding to exactly this idea that AI agents the things that we have seen the things that I get to work on at Writer for enterprise and ask AI to go do something and have it take multiple steps not having to specify these are the exact steps I want you to take but actually defining the outcome and just letting the AI go do it. So, I think to me that has been seeing that in the consumer setting is actually amazing and and more people getting to feel that magic. And I'll tell you my instinct story in just a minute. And actually, let me get into the instinct story. I don't know what instinct influencers instinct.

5:54 >> I was going to call it I was going to call it an invertebrate, but I don't think that's very >> That doesn't even work either. That's a tough one. But what it did for me, so I'm going to St. Louis this weekend. But >> invertebrates. Okay. Actually, all right. We're going to stick with that one then. Now we're forced to stick instinct inver agent. I don't know. Okay. We'll >> tell your story. I'll workshop the >> workshop while I'm going. Yeah. So, so I am going to St. Louis, actually leaving in a few hours for a family event. I had booked on Booking.com a Hilton a Hilton hotel for the weekend. my uncle, I totally missed the email, sent a fam email with a family block code in it. Instinct actually without me asking it to do this, suddenly just pops up and is like texts me, "Oh, by the way, I just saw your uncle sent a family block.

6:52 It's actually going to be $80 cheaper. You have 24 hours left to cancel your Booking.com booking, and do you want me to go do this? Give me your credit card," which I did. and it literally just went ch cancelled the original booking, created the new one. That proactivity is something that even in most enterprise AI like does hasn't existed yet because and I think in enterprise the bar is always going to be higher where it's actually like go take the action without me even specifying or asking it. But to me, that was this really magical moment cuz I have gotten to see for months now go do multi-step complex work and get a good output. I'm like pretty used to that right now. But this agent going and without me even thinking about it, actually coming back and saying, "Look at I can save you some money right now. Do you want me to do it?" And then going and doing it after I say yes. That was big. Like I have not had that level of experience with AI to date.

7:55 >> And you know, we talked in the past about how travel is a good benchmark because it the cost for it getting something wrong is it can be big, but it's usually manageable, right? It's not going to ruin your life. >> And so like handing it these tasks is actually really good. >> I know. I know. For longtime listeners, you know, I have often ranted about how everyone whenever they talk about Agentic uses travel as the kind of benchmark or example and I have said I don't like that and here I am. But you're right. I I like your thesis that travel yeah there is consequences but actually relative to other things it's actually slightly lower. It can always be fixed as necessary. so yeah, here I am having a magic travel agentic moment.

8:43 >> And once it proves itself on travel, then it can, you know, you sort of start to trust it with other things. Now, you know, we do promise nuance in this show, so I'm going to pour a little bit of cold water over you know, this idea that you know, Muse is this overnight success. It's not really used by that many people. And Meta has been doing a fullcourt push to try to get it in the hands of just about all the three billion of its of its users. You know, pushing on WhatsApp, pushing on Instagram, doing the most most unhinged crazy campaign on X at the hands of Alexander Weighing, its head of AI.

9:20 >> I don't know if that's a campaign. I think that's just Alexander. So, we're going to put a because I I do have a >> Okay. But let's talk about at least usage. And I'm curious what you think about these numbers. The information says Meta's new personal AI agent, Muse, had more than 500,000 people try it out roughly a week after launch, including more than 250,000 daily active users. Users have also submitted more than 2 million prompts. that doesn't sound like a lot to me given the scale of Meta. Meta has been pushing this to three billion people or maybe a good section of them. 500,000 people try it and two million prompts. That's four prompts per person. two, you know, 250,000 daily active users feels pretty small to me. Actually, it makes me kind of surprised that it's at the top of the app store given that the app store given that has like yeah so many fivestar reviews. Like it looks like people really do love it. So they've built a good product, but it's not taken off in a crazy way. and also that the the DA use like the daily active users. Okay, actually this is later more recent data from similar web saying that it's reached above 600,000 you know approaching 800,000 but still under a million. I don't know maybe I've become so you know desensitized to these things that like I think the number should be bigger u but what do you think about this? Is it basically my question is is it a niche use case or is this just the very beginning here? I it's that is the question, right? And I mean, yeah, I I don't know. I like I really thought about it because I'm sitting here. I'm like freaking out.

11:00 This is amazing, but is this actually going to catch on? And you're right, like Meta on Instagram, it's showing up right at the top WhatsApp, they're trying to push it. I mean, they Meta is able to control human behavior at global scale better than any other system or service in the history of mankind. So they should be able to crank out better numbers. But maybe to their the generous explanation is they recognize that this is a product under development and that pushing it too hard within their app ecosystem actually could be a mistake because if people's first experiences are not good then it's not going to last. It's not going to stick that that converting people to personal agend personal agents is a big hurdle. So, first let X take get Muse pill, let Alexander Wang go ham, let all these kind of things happen within us, our listeners probably, but not moving out to like just my mom and everyone else. I think maybe they are being kind of methodical about it.

12:08 >> Maybe, but it doesn't really seem like it. It seems like they're going all out for it. fact that you know at connect this week which is Meta's developer conference you know Mark or their VR conference they all blend together Mark Zuckerberg said it's going to muse will be the centerpiece of the company so I think they they really think they have something here which leads me to this question which I think we have to reckon with we've basically written off Meta's AI effort for months now is Meta back >> oh I mean I I was trying to find the transcript I think it was in April we were half joking But we were like what we were going through like who's on top of the AI world and I remember we both said is there this kind of off chance that by the fall Mark Zuckerberg will be king of the AI world will be American flag on the hydrooil Mark Zuckerberg again. Yes. And >> yes those are your words. Exactly.

13:03 >> Yeah. Yeah. Yeah. No. No. I know. But but I I admit it was a bit sarcastic and ingest, but I think Oh, they are they are so back. They are I mean they are more back than a lot of people the Amazon of the world have ever even been in the game. So yes, 100%. Do you >> I mean he's right or dis >> Yeah. Yeah. Well, I think, you know, it's interesting because you know, if if you assume that AI will continue to progress and that there will be continue to be more demand for AI products, then Meta is in kind of a can't lose place.

13:37 And I can't believe I'm actually saying this. If Meta's products fail, Meta is going to have a boatload of data centers, multiple boatloads of data centers, to lease out to OpenAI and Anthropic, you know, to help them meet their demand so it can become an infrastructure player. And in fact, you you'll recall before Muse, a lot of people were calling for Meta to basically give up on developing AI products and just become, you know, an AI hyperscaler and it still has that ability in its back pocket and that could be a very good business. And so if this fails, Meta has that to fall back on. Good business. So Meta now has an AI business. If this succeeds, Meta has a crazy business because if you think about the way that you make money from use, oh boy, it might be one of the best businesses ever created on the internet.

14:24 First of all, there's subscription, right? So it for the heavy users, Meta is going to charge them to to have extra Muse usage. So you're going to get paid directly by the users. Then there's advertising, right? The intent signals in this thing are off the charts. Like you're having Muse go out and do stuff for you, right? you you for advertisers you want to get as close to the action as you can and Muse is the action. So there's a possibility to charge, you know, pretty high premiums on ad rates.

14:53 And then finally, you know, there's there's the fact that Meta, you know, with all and we'll talk about the connectors. It's connecting to everything. So Meta becomes a demand channel for things like Expedia and Spotify. And you know what it says? All right, you want me to funnel the, you know, the billions of Muse users demand to you, we're going to have to take a little bit of a transaction fee for every time someone completes a purchase on your site. So this is an insane business in the making if it takes off.

15:20 But again, just to, you know, caveat, >> we're far from that, but it's definitely got great potential if it works and if it doesn't work. >> Okay, so I came into this conversation and I actually do believe this is the beginning of agentic commerce that kind of everyone has been thinking about or talking about. Again, this is like square in my world and it's been this ongoing debate. Will people use agents to buy things? Whether it's autonomous or whether it's helping guide you through the entire shopping process, it's done. Like it's already this is the experience already asking it to go search multiple sites. Here is what's how something fits me and go find other comparable things. And if I want to give it my credit card, which I have given instinct, I have not given meta yet. And we'll get into that around connectors and trust. but I do think this is the beginning of agentic commerce and like people shopping in this environment and this me this this UI more. What I do still wonder about is the advertising because I agree this just feels like oh my god like they are going to advertise the out of this platform and like you're right super high intent. They already have the greatest ad infrastructure and ecosystem in the world. like there this is going to be gold but no one has cracked advertising in chatbased UI AI yet like chat GBT has certainly given a little tries here and there Google >> wouldn't you trust Meta to be the one that cracks it like of all the companies wouldn't Meta be the one that has the best opportunity or the best chance >> yes yeah 100% okay actually it is amazing to me to their credit Like everyone I know has completely dissolved the bare the line between an advertisement and a post on Instagram.

17:17 Like it there's no there's zero understanding or care. Was this an ad or was this because to the credit it is so onpoint personalized relevant what you want to see that >> and good >> and the creative is good. The creative is very good. Okay. So you're Yes. If anyone can crack it, surely it's meta over chat GPT. Google. Oh, and I definitely want to get into where the where is Google on all this, but but yes. Yes. Okay, you're right. You got me. Meta.

17:50 >> So, there is the free tier. they there's the power plan, which is $20 per month for 500 million Muse tokens delivered weekly. And then there's the max plan, which is 100 million per month for three billion Muse tokens delivered weekly. Okay, let's >> I don't know how that will be adopted or won't be adopted, but at least it's there. >> Well, actually, and maybe it is almost a check against people basically if they're giving this out completely for free, people just moving over all their work in other systems and just melting tokens within Muse and Meta and having them pay for it. Maybe that's it's a kind of a hedge against that, but I I still don't think that's going to be the the the main value for meta in this.

18:39 I think it might be a nice to have, but I think >> they they have to crack this is advertising, this is commerce >> or other actions, lead genen, etc. But do you think >> look three spokes so three different potential businesses? Go ahead. Well, actually, no, no, go, go, because then I want to get into the trust and connectors side of it. >> Okay. So, so I think at this point, the fact that we haven't talked about open AI, is is wild. because as much as this is a meta story, this is an open AI story. And I thought about you a lot as I've been, you know, trying to figure out what's going on here. How has Meta been able to do this? and you know you've said in the past that OpenAI had made a mistake by doubling down on enterprise and what it really should have done is take the fact that it has this consumer product that people really like which is Jad GPT which has a billion users and you know keep basically hammering this home and trying to own that consumer experience and they got beat here by Meta. They did. Now I expect them to answer and I think it'll be a popular answer and I think people will trust that a lot more. but this is this is I think as much an open AI story as it is a meta story.

19:56 >> Okay. No, no, you're right. Like they had the moment and again I want to get into Google too because where the hell are you Google? None of this works without access to Gmail and Google calendar and stuff like that. But I think OpenAI actually they will respond. They will clearly I mean I think there's been a lot of like illusion to the idea that their personal agent that's going to do this kind of thing is going to be coming out soon but like you're right they doubled down on the enterprise their chief revenue officer or Denise Dresser left after 7 months they've had kind of total turmoil on that side. and did they lose a consumer? And again, when Meta comes at the consumer, God help whatever company is trying to get in their way. So, I think you're right. This is especially ahead of the IPO and kind of coming up with a clear story. This cuts so heavily into any kind of chat GPT consumer story that if they're going to fall back on that, this is going to really cut into that, >> right? And you would think that since OpenAI has better models and that's undisputed that they'd be able to come back in and challenge Meta in a in a serious way and build a better product.

21:15 But and I think we're going to get to this in a bit. You know, that is not guaranteed. And this this release has showed us that. So, so hold that thought for a moment because, you know, I think before we move past Meta, we need to talk about the marketing that Alexander Wang, the head of Meta's AI lab, or or the super intelligence lab that they have over there, you know, in Menlo Park, we need to talk about the marketing he's been doing. And the marketing has been unhinged. And, mom, if you're listening, probably a good time to to hit pause and go do something else because I titled this section butthole marketing. and and it kind of all began where somebody created a chart of AI logos that don't look like buttholes. And it has the Gro logo, the Gemini logo, the DeepMind logo, the Microsoft Copilot logo, the Claude logo, and the Chat GPT logo. And all of them get a zero because in this user's imagination all these logos do look like buttholes. But muse is a squiggle and so it gets a one because it does not look like one. And and Alexander Wang has really really run with this. he's he's tweeted it and said first of all he said state-of-the-art Then he tweeted it and with the logo and said you cannot a you cannot out anti-butold me which is like this is the leader of a major AI lab. that is interesting. Then it got it got even more disturbing, shall we say, because you know the Muse mascot, which is now beloved on the internet, kind of looks like a fluffy labu. I think that's the best way to put it. And what Alexander Wang has done and encouraged the internet to do, he has sexualized the labu, the the muse labu as is as this butthole discourse has taken taken off. So, for instance, he put a picture up of, you know, the Muse mascot doing that famous Kim Kardashian magazine cover where she sprays champagne and it arcs over her and splashes off of her butt. He did a picture of of he he amplified a picture of himself and Mark Zuckerberg and the Muse mascot all showing their butts to the camera.

23:35 and then Ronjan you seized on this they they announced a partnership with Box with a normal graphic and Alexander Wang the head of Meta's AI lab wrote weird that this is the graphic that the partnerships team told me we were going with. and it is a picture of the Muse Laboo leaning over a box box and flirtatiously showing its butt to the camera. And I can't believe I have to do this, but this is important.

24:06 >> Keep going. This is butthole discourse and reporting on butthole discourse is why people come to the big technology podcast. >> Okay. And so I in my naive sense tweeted this and said, "Why are they sexualizing Laboo?" And and I think I have an answer. >> What is your What is your answer? What is your what's the what's the theory here? This is my my real investigative reporting moment here on Big Technology Podcast. Think about what's been going on in all the other AI labs.

24:40 How they've told you AI is going to take your job. AI is going to kill you. AI is a imminent threat. AI will escape. If you're coming out and trying to position yourself against those labs, what would you want to do? You'd want to have the marketing as harm as far away from that as possible. And nothing could be farther away from AI may kill you than the Muse Laboo mascot leaning over an enterprise AI logo with heart emojis showing its butt to the camera. And I think this is what's going on. They are intentionally going as silly and as absurd as they can to show you that they're not going to participate in the end of the world discourse and there's nothing to fear here. And I think it might actually be effective sort of anti- consensus marketing.

25:37 >> Okay. Okay. See, I agree overall and I do think like even the usage of that character and I I had actually before Alexander Wang started going off like that. I had said something about the character feels unsettling almost because it's too cute and like you just feel something's going to be off and certainly things would are off. But okay, I like this idea and I do think overall that is the idea of distract and move away from human extinction into just kind of absurd and cute. My thing is so to Meta's credit the one of the more and even the way instinct speaks to you, the way Muse speaks to you, they really nailed the voice like it's like not Claude. I mean, Claude, the last couple of models, just speaks to you like it's like some obnoxious PhD professor. I don't know.

26:32 Like, what is the most like obnoxious type of personality you can actually listen to and that's the way it speaks to you? Versus, I mean, Muse, it's literally like, "Oh, yeah. Hey, quick heads up. Your flight just got a notification. It might like it's literally someone just kind of chatting with you." But to their credit, too, I actually was like, "Oh, how long for me to drive from my house to the the airport?" and it actually was like, I don't Hey, I don't have your address.

26:58 That's okay, though. Maybe do you want to just give me your cross streets and that'll help. Like, they know people are they've built this around people being a little bit leerary of giving Facebook your data, your life. And I I've connected my Gmail to Muse. I've not connected any kind of credit card instinct. I've given credit card info. I'm still like trying to decide what my comfort level is overall on this. But what I don't get is they have to build trust. The more you give to these systems, the better they will perform, the more valuable they will be.

27:35 >> If the if the like how old is Alexander Wang? 256, something like that. Like if you have a 25year-old multi-billionaire hypersexualizing a little lab boooo and then you're looking at it and you're some kind of normie which and probably they're just not even seeing that I guess but like if that becomes more of a story this is the person you are going to give your life away to and trust to have AI go do stuff for you. I don't I I think that's a bad play. I don't think that's if that's the distraction from human human extinction.

28:12 Just keep it cute. The tomagoi muse charm Zuckerberg. That stuff's good. But like this is a little creepy and like I don't know. Do you don't think it's a little does it build trust? >> I'm not saying it's going to work. I'm not saying it's necessarily smart, but I am saying that when you're talking, you know, you you said like, all right, you're talking about the end of the world, you're not talking about failing businesses. I think when you're looking at these absurd memes that Wang is tweeting, and by the way, he's 29, you're not looking at you're not thinking about the existential threat that AI can lead to. It's sort of the it is the opposite of that. So, I think that is the strategy. I don't know if it's going to work, but I certainly think that in the back of their minds, they're thinking, you know, the more absurd we can be, the more we're sort of poking at this end of the world narrative. See, I I'm still >> working against it >> now that No, no, the more we're talking about this, even kind of that argument, I I'm asking that a to kind of try to take a bit of control over this and I hate traditional corporate PR communications type things, but like like again going back to just you are asking for the trust of people. If people, the average person who is not kind of like a poster 20some male on X is going to start hearing about or seeing this kind of stuff. It's still I I have to imagine that would make people much more uncomfortable cuz just thinking about like it it actually moves the conversation away from agents are going to lead to human extinction back to the much more mundane but real oh wait weird tech people are going to take all my data and use it to monetize themselves and give away and like steal my privacy which is the worst conversation Meta ever wants to be in.

30:14 So, you got a good product. Let it run. Let it run. >> You know, we do we do talk a lot about marketing on this show and maybe just a lot of it is is bad marketing. So, you mentioned the Tamagotchi. We should talk about it because it's not only the agent, there is a device as well. Looks like a Tamagotchi. This is from Bloomberg. Meta debut's dedicated charm device for using Muse Aai. Meta Platforms unveiled a palmsiz dedicated gadget for using Muse, the company's popular new artificial intelligence assistant. pushing deeper into the AI devices market with a surprise announcement. The product was born out of the company's new design lab led by Apple's former interface design chief Alan D and it super intelligence AI group. It's called the Muse Charm. It looks like a Tamagotchi, the digital pet from the '9s. It has a roughly 2-in display and runs an operating system designed to bring Muse to life on its own hardware. it's also it is kind of cute, right? You just kind of hold it and there's like a little Muse avatar there and I guess it can get stuff done for you. I don't know what's your reaction to this. I I tweeted about it and I was like, well, what do you what well who's going to be the brave one in their social circle to break out the Muse Muse Buddy first? And it it certainly like it's some of these devices like Apple Vision Pro for instance, you know, they seem cool in the demo and and they do cool things, but there is a weird social element to it. The same thing that had the Google glasses. I don't know if you're sitting at a table and somebody brings out breaks out the Muse Buddy and asks it to like call it an Uber. I don't know. Are they are they the cool person in the group or do they not get an invite to dinner the next time? What do you think? Well, also what is the ven diagram between people who like cute tam tamagotchi type stuff and and early adopters of personal AI agentic hardware maybe maybe it's we'll find out there's some overlap we will find out but so okay >> in terms of meta being back I have to say so the one any experimentation on form factors of hardware I am excited by even the rabbit R1 rest in peace the humane pin. We talked about all these a lot. Someone is going to crack a new form factor. Anything that moves us away from just looking at our phones that are rectangular blocks and maybe foldable rectangular blocks. But still, I'm excited. So, I I I love that Meta is trying. But even bigger to me, actually, I think the audio only glasses.

32:48 This is really interesting. I like I actually was close to putting in the pre-order. I might go put it in this weekend. so I have the Metray bands. You do too. We've talked about it a lot. They were great for photos and video, especially like on vacation, walking around, walking around New York City. They were pretty good at they were good at listening to audio, but Meta AI was effectively useless on them. Now, the idea, so obviously there's been plenty of backlash in in fairness to even though I do love the camera and I use it responsibly of people being creeped out by the camera, if Muse can translate into an audio only interface, well, for how good it is on the phone, I mean, audio only AI glasses become huge. Like I I still can't even start to think about if that becomes a form factor that you're just walking along again that proactivity that you're walking along suddenly it's like oh wait American Airlines just sent an email that because of the storms there might be a delay that's massive and to have that directly on your face with no creepiness of a camera. So to me that's the biggest part of the whole hardware announcement separate the VR glasses and stuff we can get into but I think audio only Ray-B bands is big.

34:10 >> Yeah, I think it is. It is and they've been called the pervert glasses and I think they need to reverse that narrative. So that's one step to do that. Are it's okay. Are you going to buy that the cameraree glasses and are you going to buy a Muse Tamagotchi? >> I will buy the cameraree glasses audio only. I think I I think I have to I'm committing to that now. No, I'm not going to get to Tomochi. I'm like I don't know. That's That's too much. And And again, I can never look at that thing the same after what Alexander Wang did to that poor little Lubu.

34:45 >> It's like it's weird. It's weird. It's like you can't like you can't only look at it. Oh god. >> Just leave it alone. Alexander way. >> Well, I'm glad I'm glad we spoke about this. yeah, >> strong emotions on your Someone tweeted, Tom got your data. I thought that was funny. >> I mean, they have to I can't they have to not fall down that path of like this is one big data stealing story. That's the only risk they have to leading in this race. Like otherwise, they they're going to push this to actually do do you know what this made me think about though? On one hand, their distribution advantage makes me convinced. Obviously, they're going to win. They're just going to push this out to everyone. But I it Have you been on Threads recently? Cuz that was their last create a product that should be dominant. And have you been on threads at all or >> No, I'm only on threads when Instagram like basically makes me go there. But the incestant notifications for me to be on threads.

35:56 Yeah. Now I I was wondering like that actually showed me that you can will muse end up like threads where they have a distribution advantage. They're going to get some number of people who are interested in that kind of it's the same reason Twitter never actually grew or X in terms of overall user numbers. It's because there's only weirdos like us that love to spend time on it and like threads the TAM was not as big. So yeah, that's what I was definitely wondering what is that an ominous potential signal for where Muse ends up.

36:36 >> Well, I don't know. I mean, basically the question is the competition like with Threads like Threads is bigger than X, but it's also like the the size of X is not what matters. is the influence and it's hard to argue that threads is 120th as influential as X. So I'm not 100% sure. but I think we should you know what the the interesting thing here for me is that Meta was able to build this Muse product with an inferior model and what does it mean for OpenAI's ability to compete with it even though like OpenAI has better models?

37:10 do you know are those better models going to make a difference? And you know, going back to what we spoke about last week, if you can build good enough or good AI products with standard models, do you really need the frontier? So, we're going to talk about that when we come back after the break. Before we go to the break, a word of thanks to you, our audience. last week, we talked a little bit about how the pedants were owning our reviews. And of course like when we're going out to try to interview somebody like Jeff Hinton or Deis Sabis and like the you know the front page of the show is a bunch of like one-star pedantic complaints it becomes harder to have you know to to book these interviews and of course like the show gets worse. So, we knew that there was a lot of people who who liked the show and just asked you to express it and boy did you answer the call last week and u you know it takes about like 30 five star reviews to wash out one and and it looks like we got that and more. So, you're the best audience in the world. Really believe that and just so thrilled that you responded the way you did after we asked for some fivestar reviews last week. So, thank you. I'll leave it alone for as long as I can. and and just appreciate you all for doing that. All right, we'll be back right after this. And we're back here on Big Technology Podcast Friday edition. joined as always by Ran John Roy of Margins. All right, Ranjan.

38:29 so, so before we get into the fact that or maybe maybe this all connects, right? I was going to say before we get into the fact that Meta was able to build Muse with you know an an inferior model to what Open AI has but but and they have and so we'll talk about the implications there but one of the interesting things that came through my feed just to go all harness hive here was somebody tweeted it's it's not the harness that it's the harness that matters not the model with the picture of like the guy in the old oldtime meeting trying to stand up and make a point. And instead of the guy's face, there's the muse labu face there, which is it's G-rated, so we can show it. but but the thing that's interesting is what Meta did here was they took a good enough model, right?

39:21 Because we've seen all the benchmarks. We're like the meta models and the Grock models were chasing the frontier at a lower cost and we're like, "All right, let's see if that takes on." But what they were able to do is use that model which was not the Frontier model and then hook it in to all these different services via connectors and that has in turn made a very useful product without having to use Frontier AI intelligence.

39:46 So you were right. It was all about the harness and they're doing work. So talk a little bit about you know what you think that says about the future of AI and how important these connectors are. I'm glad we are saying this. I I the only my only quip with that tweet is that's kind of the confess your unpopular opinion meme framework. And it's not an unpopular opinion. I've been saying this, others have been saying this. And again, I have seen this firsthand at Writer in terms of the like our own foundation models that are not at the frontier, but are advanced, but the harness and the product and how everything connects to other systems data for a while now. So, I think I've I've had an advantage in terms of why that should work and what we're seeing right now. I But maybe to try to be nuanced about it, there's two questions right here. Do you need that level of connectivity, tool definition, overall data, how you handle memory, compute, like to make muse work? The frontier model argument has always been that still over time in like frontier models will be subsume all of that infrastructure and just be able to do it without most of that infrastructure. So, I think I don't know. It's still relatively open, but I'm very glad that like it is like no one is talking about what models it's running on. Muse is running on and I love that we're only talking about not even the harness is good. It's the product in the end. So, I'm I'm have always been harnessive team product and it's finally getting some recognition.

41:32 >> So, it's a big win for for Ron John. Now I would argue the models have the the models have gotten to the point where this is possible and therefore they are able to use these harnesses and connectors. you know, we should talk a little bit about the connectors because basically, you know, they have been tweeting a long list of companies that are going to connect with Muse, things like Best Buy and Gap and Fanatics and Walmart, Sephora, Dick Sporting Goods, Wayfair. I've mentioned before, Expedia, Shopify, has connected with it. So, talk a little bit about what added functionality that that I think T-Mobile, I used it with T-Mobile this week to check on my phone bill. why are what are these what do these connectors do and why are they important?

42:16 >> Yeah, I think it is really important on two sides. There's the agentic commerce side. Again, this is where announcing all these like partnerships means these companies are going to be pushing direct feeds into Muse rather than it having to go use web search like computer use to try to log in to try to get to your find different items. So I think that's going to be very interesting. Trying to replicate the activity that would take place in a dedicated iOS app but directly within the Muse app and ecosystem is very interesting. I think there is something too that it is still clunky. It's gotten a lot better of logging into T-Mobile via some kind of web browser takeover. Now being able to do that kind of work I think gets very interesting. But to me the most interesting tension here is and there was a lot of chatter around this this week. It's so much of so many businesses is built around consumer lack of inertia. like customer I mean T-Mobile I'm sure like do they want to have you be able to go in and be like actually explain to me that five-page PDF phone bill that I've never quite understood and never took the time to upload into chat GPT and go consu comparison shop and tell me how I can save money and get find a better plan like this is gonna be re there's a I mean this is what a lot of people like Expedia Expedia does not actually own the inventory. Expedia is a price comparison engine that then will guide lead out or and make money off the lead genen to the travel site. So like does that start to go away? It's actually interesting to me that these companies are jumping into these partnerships because I actually think this disintermediates a lot of existing businesses. This actually changes a lot of businesses that have been built off people forgetting to unsubscribe or whatever else. So to me, these partnerships, I get it. You're trying to look cool and add it to some Magentic Commerce press earnings call or something. I don't know. But I still think it's not good for a lot of these businesses.

44:34 >> Yeah. Yeah. And it was interesting because there was initial like joy, you know, when this came out and when these companies were connecting with Muse and the stock market really reacted to it like they sent Shopify up, they sent I think Instacart up, they sent Expedia up and then all these companies maybe outside of Shopify gave up their gains like immediately. And there's a chart that's been making way its way around X called the Muse selloff travel aggregators versus Meta. And you've seen Meta up, you know, 20% since, August 25th. and who's down? It's Uber, Door Dash, Expedia, Trip Advisor, Instacart, Airbnb, and the list goes on. And it's almost proportional to the gains that Meta has had, the losses that these companies had, and I think that's actually, you know, I was thinking this week, you know, do you give these companies credit for playing ball with with Muse? because it's going to basically like even if Muse doesn't take off, this is going to be a thing most likely and so they're going to make the decisions they make now will will definitely influence how they operate in the future. So do you give them credit or do you say you know they're probably toast and I I I agree with you. So I think the rise of you know apps like Muse you know they become they become threats to these these platforms because they basically perform like a app like Muse will perform the function that something like Expedia did where Expedia used to be the agent where it would go to every like travel site and find you the hotel or find you the flight. Now Muse can do that. So it sort of makes it pointless. I I think this plays out like the death of SAS where nothing's going to happen immediately or dramatically.

46:10 But I do think there's a strong parallel between the idea that like enterprise agentic AI is going to kill off SAS which I still believe over time. I think this does any kind of like non-inventoryowning marketplace sorry not like effectively agent as you said is severely under threat because and if it's not muse it could be instinct it could be google releases some there's like all those services whoever wins in that race or if there's multiple players they do the same work that the expedias of the world have done but I mean yeah commerce is definitely gonna change. I think that's that's a fair thing to to say.

46:55 >> Okay. So, before we move on from this, I need to ask you, what does this say about Frontier Labs in general? Right. Because if Meta was able to build Muse with a non-frontier model with the harness and connectors, do you need Frontier Labs to do most things? And it goes back to our discussion from last week about the the chart we brought up from the ramp economics lab where Frontier model usage was 53% of all AI spend August 2nd. It dropped down to 45% of all AI spend September 6. Now it's gone up and down right it's had its moments but the thing that I found most interesting was this is from Eric Karazian who's the chief economist at RAMP. He says if you see this chart know this is my perspective. The revenue going grow the re the revenue growth path for openai and anthropic has weakened due to the competition and that competition is from other US models not China and open source. So the standard models are doing a good enough job and the standard models are often cheaper.

47:56 It's a trend across multiple data points >> and it's a massive threat into any kind of IPO story and I think that's what I'm not saying directly the human extinction conversation has helped distract from but it is a massive threat. I firmly believe, I've seen this firsthand, like the conversation change. If you want to do 3D world modeling, Astra's your guy. Like, great. If you want to >> everything else, go to Alexander Wang and the sexualized labu.

48:29 >> Yeah. Go to the labu. Oh god. Sorry. >> I need to just erase that from my mind. >> Have to do but it's not it's not forever. The laboos themselves. I can never have one in my house again. not that we have them. We don't have any Laboos, but I can never have a Lubu. Yeah. Well, >> wait. You have a throw ours out. Yeah, we have Lubu keychain. >> Get rid of that thing. Or look at it.

48:54 Look at it and tell me how you feel next time. >> Feel dirty. I I think like this is going to be to me actually the single most important conversation of the fall in the next few months because why pay for Frontier model? Can it other models do the job? What is the added benefit? Does something about Frontier models like subsume all of the harness? This is the debate like and it going into this kind of IPO environment like it's a dangerous one for these companies.

49:33 >> Yeah. No, I I have a story coming out about it on big technology. It's called right now the working headline and I'm going to publish it later this afternoon on Friday metas muse revival in the AI frontiers unspoken threat. I think it's a big deal. >> No sexualized labu images though please. >> No I will not. Only for the people watching YouTube will I put those video those images up. Everybody else or I guess on our shorts everybody else is freed from them. all right before we go got to talk to you about dopamine sites and we can do this briefly.

50:03 there is a phenomenon where people have gotten so much dopamine from like shopping online that there are these new sites called dopamine sites that are meant just to mimic that activity and you don't actually do the activity. Right? So this is this is from one story about it. This week I placed orders for a $44,000 PC Philippe hand engraved watch, a $12,000 500 $12,500 Hermes bag, $9,800 Tiffany diamond ring, and a $7,000 Cardier love bracelet and yellow gold. I don't need any of them. I certainly can't afford them. Thankfully, they'll never arrive. Instead of Amazon, I spent the past week browsing a new breed of websites called dopamine sites.

50:50 A trend that's emerged in South Korea. It's called like dopamine shop or food never comes where you order food and shop online. and you just do it for the dopamine hit, not for the actual products themselves. I think we just have to talk about this before we go. what you must have some perspective on these dopamine sites. It's crazy, right? >> So, okay, my my hot take on this is going to be I actually think this is great. I think this is Yes. I think because I was thinking about it for my initial reaction was like, "Oh my god, this represents the end of humanity. This represents like this is the most depressing thing ever." But in reality, it's just kind of like a video game at that point. It's just kind of like I mean, we all do it like scrolling Zillow and setting alerts on houses that you're not going to buy. shopping like you see these every I mean I having worked in digital commerce for many years like so much of the experience is around the thrill of the search and the dopamine resulting from that or getting that deal that is entire basis of retail in general. So something that kind of it's like a turns it into a little bit of a video game to me. kind of feels like actually dress up games are quite popular where you just like buy clothes, any kind of digital good that you're buying and now you don't even have to buy it. You just go through the shopping itself. So, >> I am pro dopamine sites and if it's just if it's not sports betting and other more negative and nefarious dopamine hits, let's Why not, Alex? Why not?

52:28 >> someone had this amazing idea. They wrote multi-billion dollar opportunity in combining these dopamine sites with the lottery. So you pay $50 and there's a one in 1,000 chance that that PC actually arrives. I love that. Could you imagine? >> Well, I mean you you know there's like bid up. I think there are these all these sites that are really kind of scammy, but basically are built around like you put in like a dollar and you might win something that's like $8,000, but then they only do it when a certain critical mass does like this I guess this model has definitely this exists in the scammy corners of the internet forever. And now but now they're not taking your money. They're just how how are they monetizing, do you think? Or they're just like >> it's just for the love of the dopamine.

53:16 meta. So they think I'm muse. >> No, you know what? They are doing it for the love of the game. The love of just giving you we'll we'll pay for a little bit of compute and you get the dopamine. So I'm pro dopamine sites. >> I can't get on board with it. But look folks, you know, if you don't want to shop on a dopamine site, you could always get your twice weekly hit of dopamine here on Big Technology Podcast, okay?

53:45 won't always be perfect. Promise that. But but we're here and we'll we'll we'll hand that dopamine over and we won't make you go through a fake sales process. >> No. >> Sounds about right. >> Give us your data. You don't have to give us your data. >> No, don't give us your data. Absolutely not. We don't need that data. >> I don't even know what I would do with it. So, >> yeah, me neither. Ranjan, great time as always. Nice to talk about something non-extinction as as you said, >> though. I don't know. sexualized laboos.

54:16 I don't know what wins in the great the great debate, but >> well, if you folks, if you've stayed this far, we appreciate you. So, thank you for doing that. And I would promise not to have any conversations about sexualized labu next week. Although that that's a promise that might be broken. >> All right, see you next week. >> All right, Ranjan, thank you again. Thanks everybody for listening and watching and we'll see you next time on Big Technology Podcast.