Muse Wonder: Meta Is Back in the AI Race | Lightwork

0:00Can you help me book a flight? Can you help me clean up my emails? Those don't require super intelligence. It requires focus. But it turns out they're not satisfied with good enough because this model is very impressive. Am I ready for it to have 24-7 ambient access into my conversations at home and pick up on things? SpaceX specifically also wants to become your phone company. I also think we're going to see the Tamagotchification event that we've seen play out kind of across these apps.

0:36we're back with another big leak meta released muse wonder its first frontier model and google rolled out gemini for argin the next day apple put a new siri into three home devices including a display that recognizes up to six faces and spacex bought the low band spectrum in its goal to become a major mobile carrier in the U.S. Plus, two updates from across the Lightspeed universe. Let's get into it. Claire, we had two new models this week, both claiming to be the frontier. We had Google release Gemini 4R gun, and Meta announced Muse Wonder and Wonderful. What stood out to you about these models?

1:19What's special about these announcements? Yeah, I think a lot of people have been sort of watching what Google and Meta would do in the space given OpenAI and Anthropics very steady pace of releases. With Meta in particular, I think this is notable because this is the first frontier model from MSL, Meta Superintelligence Labs, which was the unit started by Alexander Wang since Meta's$14 billion acquisition of scale. And a lot of people have been really curious what they would launch, especially given that, you know, back in April last year, their open weights release of Llama 4 was sort of a flop.

1:55And Zuck just re-orged the entire AI unit around bringing in Alexander Wang. And I imagine since then, it turned out to be a really good thing given the success of something like Muse. But this is the follow-up model launch on top of the product layer. And Google, I think, is in a similar situation where, I think similar to what we've seen with OpenAI launching Astra, Anthropic launching Opus 5.5, People have just been curious what is their state-of-the-art frontier model and are they falling behind just given that a lot of their more recent releases have been more product layer as opposed to model layer.

2:29Yeah, and there have been some questions on the timeline about whether these are truly frontier models. Our portfolio company, Arena, did a really great breakdown on Muse and took it through a whole bunch of paces. It's still not quite there in terms of like AI web development. It makes a website that's a little more like AI Slop compared to Opus 5.5. But what it's really focused on is the use case for Muse, which is agentic work, where it seems to do very well across the board. What was interesting to you about this model release? And, you know, do you think that it's the frontier? Does it feel like it's just branded as such?

3:11Yeah, I mean, it's tough because everyone's a frontier when they're sort of benchmarking against whatever, picking and choosing. It does seem like everybody's bench maxing, right? So it's like, it is good that we have players like Arena to at least be a single source of truth. Yeah, it's super helpful. And I think for folks who are interested in a lot of these model releases, like I've loved their, what are they called? Unboxings of models. It's been really helpful to keep up with everything happening in the ecosystem. But I guess specs wise, Wonder is actually two models. So they have Wonder, which is on the API that allows developers and any of us to plug the model directly into our software.

3:52And that is$4 per million tokens in$20 per million tokens out. That's about the same price as Claude Opus 5.5, which is funny because I think both models were benchmarking price against Opus 5.5, which sort of tells you something else as well. And then the better model actually is called Wonderful. That is actually only available inside Meta's Muse Code, which is a subscription. I think it's about$50 and$200 a month, depending on the tier. So you actually can't get access to the front tier unless you pay for Muse Code. And so MuseCode is basically their version of CloudCode or Codex, which is interesting because I think you're seeing the more impressive product locked in into their ecosystem.

4:37And I think that also tells you something a lot about how they're positioning themselves. I think they've sort of moved away from open source entirely with this. I kind of thought that they were just going to allow Anthropic and OpenAI to be the frontier and were happy that they had this incredible consumer hit that was just going to be good enough. But it turns out they're not satisfied with good enough because this model is very impressive. And to what you brought up earlier, it is specifically very good at two things. One, active memory, which is how well it can remember details about you across conversations.

5:15And two, active focus, which is how long it can keep working at a task and how persistent it is. So they're really betting that these two things are, I mean, they are largely very important for all long horizon running agents. But I think especially in Muse's case, they probably don't care as much about, you know, hitting specific state of the art, reaching physics or math benchmarks, but rather how useful is this? And I think persistent or active focus and active memory are really core parts of getting to very well working agents. Yeah, I think Zuck and Alexander Wang said this on Cheeky Pint.

5:54They are actually not as focused on the coding side, right? Yeah, I mean, I don't think your average meta user on Instagram is whipping up entire repos of code on Muse. No, they want to know, can Muse get the task long range? Can you help me book a flight? Can you help me clean up my emails? Those don't require super intelligence. It requires focus and memory. Yeah. And then moving on to Google Argon, what makes this model so special? I saw that it was able to work overnight and accomplish several tasks. Like, what are the big things that makes this model so special? Yeah, it was announced in a couple weeks ago, I believe at the end of September.

6:43And for the last few weeks, it's only been available to vetted cybersecurity partners. I think you've actually seen that pattern with a lot of the frontier models. Seems like it's the new thing you must do before you release a model. I think Fable really set that precedent. But as of a few days ago or more recently, it is now available across API, Antigravity, Gemini Enterprise, Gemini App, and AI mode in search. And specifically what it's great at, similar to, I would say, Muse, Wonder, and Wonderful, is a lot of focus, I think, on legal, business, and finance workflows. I think it's able to do really well mixing vision and code, which I found.

7:27So you can build apps, 3D scenes, animations, and I think I saw a demo of someone building a fully working music production app from a short prompt. So I think those are sort of the spaces that I've seen it excel in. Another video I saw was on how it could be applied to spatial reasoning, which is really interesting. I think someone fed it a video of someone playing Cornhole, and then it had - I saw that. It was amazing. Yeah, right, a simulation. And this isn't the same as some of the VLAs and robotics brains that we've talked about previously on the show. This is not skill. Yes. This is different.

8:04Exactly. This is not that. What it is, though, is it can now very clearly work as a planner and a simulation builder. So it understood what was happening, worked out the physics, and then wrote a plan and then implemented that. And then they actually put that into a Aptronic humanoid robot, which is a company that Google has partnered with. So I just want to make that very clear here that this is not some massive breakthrough. Not a robot brain. Yes, it is not a robotics model. But what it does show is that it has insane visual and general reasoning. Claire, Apple unveiled Home View. It's$299 entrant into a market that's been around for a very long time.

8:47I feel like we've had the Amazon Echo show in homes for years. People have the products from Google and now it's been rebranded as Gemini speakers. These things have existed for a long time. Why is Apple choosing this moment now to enter the home display and home automation space? Yeah, I think this came out of John Ternus' Welcome Home event. And I think it made a lot of sense because John Ternus is known for his hardware engineering background versus, as we talked extensively, Tim Cook has been, I would say, mostly the guy who had created Apple as the platformer as we know it today. But they launched three things.

9:29So the home view that you're referencing is$299. And it's this smart display has like the seven inch screen. I think you mentioned when we were chatting before, this is like kind of looks like an iPad attached to a camera. Yeah, it's an iPad attached to a speaker. Yeah. And it's what it's able to do is it can recognize up to six people and show each person their personalized calendar, their messages. And it only works if it sees your face and if your unlocked iPhone is nearby. and it's specifically built for homes that already use iPhones. And then they also launched a couple other things like a better Apple TV with Siri AI integrated and another HomePod.

10:09But yeah, it's interesting. I think this is Apple's big splash into the home. They did previously, I think, launch something called HomeKit in 2014, but that didn't really work. I feel like everybody's got HomeKit on their phone and you can obviously control some automations in your home that way. But this does seem like they're trying to make like the dashboard for your home. And the video was great because it made it really centric around the family's morning experience. I thought the video was tremendous. What will be interesting to me is will people choose Apple because they like the ecosystem?

10:42Will they choose it because they value Apple's stance on privacy? Certainly, like if you have an Amazon Echo Show, like it shows a lot of ads on your table. I think the Google one does a little bit less of that, but it's still not really designed for you. And this feels very much like centered around the humans who use Apple products. So it's opportunity for a wedge, but it seems like it's a wedge into a very old market. And I'm interested in how it actually plays out. Do people buy these? Is this something that becomes a huge hit? But I think the most interesting thing for me was just that I always lose my Apple TV remote.

11:24And the fact that this one actually can find its remote lost in the couch seemed like the biggest thing in this announcement. So on one hand, I think for me with the personal agents on my phone, like an Instinct or a Muse, I've actually traded off privacy because I just want the best thing that can get stuff done. You're not letting it into your text. Exactly. And so I think that's the difference is home. I actually do think something like Google or Apple, and Apple especially with such a heavy emphasis on privacy, might have a significant advantage is because I don't know if I'm ready to let instinct or clot into my home.

12:00And Muse for that matter. I think all of those are great because they're accessing my emails. They're booking flights for me. But am I ready for it to have 24-7 ambient access into my conversations at home and pick up on things? I feel like in that world, I would prefer someone like Apple. So that's actually where I would say the safety and privacy part does have a significant advantage is like the entrance into physical homes. That being said, I think all of them are probably going to go after this broader vision of collecting and being helpful in ambient intelligence. Like there was a reason why OpenAI is working on this like third core device.

12:37Yeah, this mystery apple donut puck thing that they've been teasing that they brought Johnny Ive in for. I think that's like another similar play here is how do we actually collect audio and visual intelligence to be even more useful that I can talk to you while I'm doing my groceries or looking through my fridge. All that is not unlocked right now. So I don't know, at least for me, that's my decision making framework. But who knows when, you know, if something does provide enough value, I can see myself trading off some of that privacy as well. Yeah, I think so, too. I think the interesting thing was that Apple was yet again selling Siri.

13:13I mean, it's been years and years of Apple selling Siri. Have you been able to try the new Siri? I haven't played with it too much, but I think it's past trauma. I think I'm just like, oh, this is never going to work. But it does seem that from the text to speech that it is significantly better. And that was always the hardest thing where I was finding myself like going and leaving a granola note or doing other things or whisper flow. But those always take you out of the experience. And it was just so nice to have like a one click phone button where you could do that. So I do think that that is a big upgrade, but it doesn't seem like Siri AI is yet as capable as any of the agentic tools that we're all playing with now.

13:58So we'll see whether this gets adopted in homes in a way that they expect it to. I will say, though, I do think there is a lot of room for hardware gadgets. Like I've been seeing I don't know the exact name of it, but, you know, when I've been visiting friends who have kids, they have that little skylight or wall thing that. Oh, yes. It's like a family planner. I do think that's a great example. This could take that place, for sure. Like, I don't know if it's trying to take over Instinct or Muse or any of these personal agents, but rather, can you have a more intelligent display that has some AI elements that you can talk to?

14:36But I think it's really trying to replace that. That being said, I also think these are bidirectional, though, right? Because if you do create immense value, then in turn, I want to talk to it to update it. And that, in turn, makes your ecosystem much more powerful. I would say the big player to watch here is Google and Amazon and what they'll do in reaction to this. I think Amazon clearly has Echo to what we discussed earlier. I think there are mixed reviews of Echo and Amazon's AI. Yeah, and I feel like Google kind of gave up a little bit on Google Home. Like, I call to mine, and I feel like it doesn't even remember.

15:09And I upgraded to the newest Gemini one. It's still like it almost feels like all of these frontier companies have focused so much on their apps and their other experiences that the home experience has been a little bit left, by the way. And that has given Apple, I think, just an interesting opportunity here. I also think we'll probably see a lot of maybe like pro Seamers or developers on their own building. I've been seeing a lot more of a sort of this revival around tinkering with home devices. You know, like I saw someone had basically sort of vibe coded there or pieced together a way to use audio to turn on and off their lights or, you know, really custom based off of their car entering their garage, reconfigure their home in certain settings.

16:00And so I actually think there's a lot of like more tinkery home device creation that we're seeing as well. And I actually think there's almost like an excitement of being able to take all of this intelligence and apply it to physical devices as well. I also think we're going to see the Tamagotchification event that we've seen play out kind of across these apps. I do think there's going to cross a point where it'll start to enter the home. there's like a Black Mirror episode with Miley Cyrus that kind of is like that, where it's like a little bit of like a friendly character. I do think that that's probably on the horizon.

16:36Yeah, I mean, if you make the little Tamagotchi, if you abstract that into a larger device, I could totally see us having a little orb where your muse sits in there and that's integrated with your entire muse ecosystem. Who knows? Maybe we're giving them ideas. So with this one, I ended up doing a very deep dive on how you build a carrier business, which I did not think I was going to spend a couple hours learning all that. But I mean, we've been talking about SpaceX launching a mobile carrier since way before the IPO. This is true. But now they're doing it. And we've mostly focused on the satellites part of the business.

17:11But now they're building a mobile business, too. Don't they have a partnership with T-Mobile? Like, is this competitive with that? It is indeed. So basically, I think right now, Amazon and SpaceX are all going, they're competing for Wi-Fi. But SpaceX specifically also wants to become your phone company, which now puts it sort of as this like fourth carrier alongside AT &T, T-Mobile and Verizon, which are the big three. What happened here was basically the FCC approved the 15 ,000 satellite direct to device constellation. This means that you have a bunch of SpaceX satellites that are very low orbit and that they can serve ordinary phones.

17:53Specifically, what happened here is if you think about the ways this comes from my extensive research, and I am by no means an expert on all of this. But the way that spectrum works today is that your phone talks a network over radio waves, sort of like FM radio, but at higher frequencies. And right now, you have different companies broadcasting at different frequencies, sort of like... They each have their own spectrum, right? Exactly. They have their own lanes. And so that's what the FCC approved, is they approved for them to be able to reach ordinary customer devices. But two, SpaceX paid$8 billion to acquire a bunch of low-frequency airwaves from an investment group called Grain.

18:39those frequency waves, they flipped them from T-Mobile a couple months ago. So T-Mobile owns the former spectrum that they are now going to use to launch their mobile network. Yeah. But they don't have towers yet, right? Because that's another component of this. Yeah, that's the other part of the build out. This is just very early, just the FCC approving it because this is basically regulatory speak for we want a fourth carrier in here. And I think actually after the news, there was, I believe, about a single day ratio of roughly$39 billion in combined market value. Yeah, there was quite the public markets reaction for those poor mobile carriers.

19:19But it makes sense because right now I think there's like Mint Mobile, a couple others. Right, but all of those carriers are kind of like derivative carriers and low priced options. They are not a fourth competitor and nothing is nearly as formidable as Elon Musk and the SpaceX team. Yes, definitely a very formidable challenger coming into space. I also think another big part of this is that SpaceX owns full stack, which we know that these other players don't. So SpaceX has Starlink Internet. So that's the one that we know them for, which is, you know, thousands of satellites that beam broadband.

19:58And you could see them start to create a bundle, right? Like you have your SpaceX mobile with your SpaceX Wi-Fi. And then all of a sudden, do you need some of these other things that you're already paying for? Maybe not. Yeah. And I think this is really their effort to own the customer because right now they do have Starlink mobile, but that is intermediated through T-Satellite and T-Mobile. Yeah. And if you ever tried to use that, you have to position it at the sky and it just it doesn't. Oh, OK. It doesn't work. I think at the end of the day, you need what they're doing now. You need to buy a spectrum and then you need to build an actual credible network.

20:37But I understand that that takes quite a long time. Yeah. So the other big thing here is like, do I think that overnight Verizon is dead because of this? No, not at all. And also, again, not any public investment advice. No, and they have certainly lots of other lock-ins with their customers. Exactly. And there's a lot of further buildup that this requires before Sir Link and SpaceX even become proper carriers. For starters, nothing is built yet. To your point, you need towers, you need stores. Actually, you also need phones to work with it. So every phone today has a chip that only listens to certain airwaves, sort of like radio, getting certain stations.

21:15And most phones today haven't been fully tuned in to the airwaves that SpaceX has bought. So in reaction, phone makers like Apple and Samsung will have to add support for that. Cell towers, that's a big, physical, expensive build out. But if we get there in 2030s, maybe we'll get for the first time a company that owns this full stack of spectrum satellites and the right to operate independently. And I think that's a really interesting challenger. And I think probably a better way to think about the stack and, you know, building a platform that really conveniently bridges all those different players across the stack.

21:57Yeah. And at the end of the day, like competition's great for the consumer and we haven't had a new entrant to this market in a very long time. Yeah. I feel like Internet and mobile, probably the two things in my head that have just felt like have stayed largely the same. Almost intractable. Right.

22:16Zaria, which builds real-time servicing and collateral infrastructure for structured finance and digital asset-backed lending, raised$5.2 million in seed funding led by Lightspeed. The company has applied to the OCC to charter Zaria National Trust Bank, which would put trustee, servicing and collateral management under one federally regulated counterparty. party. Zaria says its platform is already live with partners including Maple, Galaxy, Kraken, and Figure. Lightspeed is an investor in Zenzap, which is a mobile-first work chat platform where teams and AI agents share the same group chats. They came out of stealth this week with the launch of their AI agents.

22:58They have thousands of businesses across the US, UK, and Australia that use it daily. And currently, any employee can create an agent in under 90 seconds without writing code.