October 13, 2026
2026-10-13
https://www.youtube.com/watch?v=cnQizOftgt4

[-1:-1:-1] We go live in ten.
[0:13] Every morning, the future moves faster.
[0:18] The models, the money, the power, and the people building what comes next.
[0:57] From Austin, Texas to the world.
[1:04] AI News, Real Context.
[1:07] This is the Austin Wilson Show.
[1:10] I'm Austin Wilson, your human in the loop.
[1:18] Good morning, and welcome to the Austin Wilson Show where we're breaking the AI news that matters to you.
[1:25] Today, my name is Austin Wilson, and I'm coming to you live from the AI capital of the world, Austin, Texas.
[1:33] Texas.
[1:34] Uh, really excited today.
[1:35] We've got a great docket to cover.
[1:39] Some amazing, uh, news stories have dropped in the past twenty four hours.
[1:45] Today, we will be covering the latest meta model, uh, meta, uh, what is it called?
[1:53] Wonder, I believe, uh, that'll be very interesting, very impactful.
[1:57] Microsoft is betting on its own AI chip.
[2:01] Uh, Epic tells their developers, hey.
[2:04] No more handwritten code.
[2:06] Uh, small AI models may soon be able to copy themselves.
[2:10] AI chip stocks showing a little pressure, uh, in Asia, and Anthropic eyes its first Japanese data center.
[2:21] Let's get into it.
[2:24] First up, Meta just launched Muse Wonder, a new frontier AI model.
[2:29] It pitches as a real rival to Anthropic and OpenAI.
[2:34] An independent leaderboard scores it well above the company's last model.
[2:39] Developers can use it right now.
[2:42] The top version is only for paying MuseCode subscribers.
[2:47] Paying MuseCode subscribers.
[2:49] I might have to go sign up.
[2:52] Uh, a credible frontier model makes Meta a real competitor for paid AI agents, not just a follower.
[2:59] Meta says, uh, Wonder is state of the art on personal and professional agent tasks.
[3:06] Independent tester, artificial analysis ranks it well above Meta's previous model.
[3:12] Meta hasn't published its full safety report yet.
[3:16] It only says one is coming soon, which Meta has had a pretty good track record, I think, um, uh, under Zuckerberg and with Alexander Wang in showing their safety and responsibility.
[3:29] So we'll see when that comes out.
[3:31] This is, uh, another amazing development from the AI kinda model world.
[3:38] Right?
[3:38] No longer do we just have a duopoly between the the behemoths in OpenAI and Anthropic, but now we have, of course, Groq in xAI.
[3:50] We have Muse, the Muse wonder model coming from Meta, uh, and then these open source models that we talked about.
[3:57] And then last week or the week before, we talked about Google's Argon model.
[4:01] So it's...
[4:02] This is...
[4:03] The race is not over.
[4:05] Right?
[4:05] Uh, it's very interesting to watch this play out as we see more and more competitors come in, um, and, uh, and and it's not slowing down.
[4:14] And this is a really amazing thing to see because, typically, in a in a market like this, in a technological revolution, uh, that we're going through, you would see, you know, maybe one or two folks run away with it.
[4:31] And and it's...
[4:32] Especially the lead that seemingly we've seen come out of, uh, OpenAI and then Anthropic kinda take over and really take a massive lead from a revenue standpoint.
[4:43] And then their capabilities have been phenomenal.
[4:45] And and now this is really the first, uh, frontier company in Meta and with their Muse wonder model to actually surpass, it looks like, the capabilities of Anthropic and OpenAI, and this is the first such model that we have seen.
[5:04] So this is breaking news.
[5:06] This is really important to the industry.
[5:08] This is really important to, uh, society as a whole.
[5:12] When we have many, uh, competitors in here, it just means more intelligence, uh, better capabilities for all of us.
[5:21] This is also really interesting because of how fast Meta has been able to do this.
[5:28] Seemingly, they were very, very far behind coming into the summer, you know, maybe four or five months ago.
[5:36] And, uh, then they they released their latest Muse models, which were pretty good.
[5:41] And then not too much longer, we have new frontier new frontier.
[5:47] So this is very fast.
[5:49] Um, there was some speculation.
[5:51] Of course, this is speculation that, you know, we've seen a lot of advancements come out of Google in the past weeks and and months, and Google has historically been very far behind.
[6:01] There was speculation that maybe Google had found a new faster way to develop models and to improve their models through a technique called recursive self improvement.
[6:14] This is what all the labs have been working towards, which is essentially a model, uh, an AI that can train itself.
[6:22] Right?
[6:23] This would undoubtedly lead to super intelligence, and speed of development would increase even faster.
[6:31] You can't help but wonder, is the same thing happening at Meta?
[6:35] Have they already uncovered RSI?
[6:38] Are they moving at a clip that is substantially faster than what OpenAI and Anthropic are moving at?
[6:48] Uh, curious.
[6:50] What do you think about this?
[6:51] Has anybody tested out the MetaMuse personal agent, and do you think this is gonna have an impact there?
[6:57] Uh, leave your your, uh, your responses and comments below.
[7:00] Also, you can reach out to us at TAWS dot live and tell us your thoughts.
[7:08] Microsoft unveiled Maya 300, its own AI chip to power its AI services and lean less on Nvidia.
[7:17] The company says it's about 40% more efficient than top GPU systems.
[7:23] Saidiya Nadella says it rolls out at massive scale in 2027.
[7:29] Uh, every Maya chip Microsoft deploys is a GPU it doesn't have to buy from Nvidia.
[7:36] Microsoft says it does about 40% more AI work on the same power as leading GPUs.
[7:43] Its previous chip already runs parts of Teams and Copilot, and Microsoft says that one is more efficient.
[7:52] Two, this is great.
[7:54] Microsoft won't say which chip Maya 300 beats, and nobody outside has tested it.
[8:00] Wow.
[8:01] We see another software player getting into the hardware space.
[8:05] We've seen this previously at OpenAI developing their jalapeno chip, and now we've got Microsoft, uh, developing their Maya 300 chip.
[8:15] That being said, um, Microsoft has, of course, been in the hardware game for a long time, but getting into chip manufacturing, uh, especially from a GPU standpoint and these AI chips is, uh, a big step forward.
[8:28] Again, we continue to talk about the full stack of the AI workload, right, where we've got the infrastructure at at the bottom layer, the base layer.
[8:38] That's the compute.
[8:39] That's the data centers.
[8:40] That's the chips.
[8:41] Right?
[8:42] Then in the middle layer, we've got our models.
[8:44] That's the intelligence layer.
[8:45] And then on the top layer, we have our applications that, uh, you and I are actually interfacing with.
[8:51] And so we're seeing...
[8:53] Typically, Microsoft is going to be operating more in the application layer.
[8:58] Think about Microsoft three sixty or their Copilot AI, uh, tool.
[9:04] Then they've got some models that they've been working on.
[9:07] We've talked about that a little bit.
[9:08] They've released, uh, one last week, I believe, that we covered.
[9:12] Uh, and they, um, also have a long term relationship with OpenAI.
[9:19] And so they have some intelligence layer.
[9:21] Now we're seeing them move into actually the infrastructure layer and the the the the chip layer.
[9:27] So trying to own the whole stack.
[9:30] We're seeing this across the board.
[9:31] The important thing here is I don't think any of these companies are going to gain a monopoly across the full stack, which is really important, of course.
[9:42] We we want to see diversification across different, uh, providers at each layer of the stack.
[9:50] That that being said, it is great to see more competition in that space.
[9:55] And what this will probably mean, and I think what we've seen to to some extent with these, uh, kinda unique chip developers is they build the chip for a specific use case or specific software.
[10:08] So in Microsoft's case, this chip probably runs their Copilot and their specific workloads really, really efficiently.
[10:17] It's built specifically to do so, but it's probably not gonna work quite as well at...
[10:22] On a general standpoint.
[10:24] Obviously, that's speculation, but but we do think that that's where we've seen some of this historically going is, uh, they're called ASICs, application specific chips that work on, uh, the...
[10:38] Those unique workloads.
[10:40] So this is great.
[10:41] This is a great development.
[10:43] You guys think that Nvidia is going to, you know, wobble falter at all from Microsoft getting in the game?
[10:53] Let us know your thoughts on the subject.
[10:58] Epic Games reportedly delayed its Marvel and Star Wars shooter games and told the team to go AI first.
[11:06] Bloomberg says a director wrote, quote, unquote, no more handwritten code with all art generated with full AI workloads.
[11:16] Staff were pointed to tools like Anthropic's Claude.
[11:19] Tim Sweeney posted that a huge pivot is underway in the video game space.
[11:26] A major studio is betting a flagship Disney game on machine written code and art.
[11:33] According to Bloomberg, the team was told to go AI first.
[11:37] Epic says its developers lead the AI workspace.
[11:42] Bloomberg reviewed the internal message, and Sweeney says the goal is more and better work, not fewer employees.
[11:48] Right?
[11:49] Some staff reportedly reacted with shock, and the game has, uh, no new release date.
[11:56] Wow.
[11:57] This is a big development, of course, coming on the heels of what we saw out of Google last week with their announcement where you can...
[12:05] Anybody can build a video game, uh, text based first, and we're seeing this competition come, uh, into the the mainstream and the major video game developers with, uh, this announcement out of, yeah, out of our our team here, uh, on on the video game front.
[12:27] So what...
[12:27] What's interesting here is, uh, where does this go?
[12:31] Does this have implications into other industries?
[12:35] And are we going to see more folks saying, hey.
[12:39] You have to use AI to build our products.
[12:42] Epic Games, of course, has, uh, been around for a really long time, and they've been a leader in the video game space.
[12:50] But are we going to see other developers, other development companies say, hey.
[12:56] You have to use AI.
[12:58] And especially when you have competition like this, where if the competition can move faster, more efficiently develop a better product faster with, uh, AI tools, uh, and and maybe at less expense?
[13:11] Are we going to see that proliferate through the market?
[13:15] It'll be interesting to follow this.
[13:18] Uh, not only...
[13:20] You know, we haven't seen this in software companies necessarily where they're mandating that, uh, there be zero handwritten code.
[13:28] Right?
[13:29] Um, we we haven't seen that at a large scale at least, and and nobody's come out and officially said that.
[13:36] That being said, I think there are a lot of companies where maybe 90% of the software development is done by AI and then reviewed by humans before it's actually pushed into production, um, are we going to start seeing companies say that, hey.
[13:52] You cannot write any code by hand.
[13:55] And and what does that do to the development space?
[13:58] I think it's actually really interesting because now when when you're not focused on the technical aspect and the engineering aspect, which, of course, engineering is is a process of building things, testing them, breaking, trying to figure out break fixes, and then do that iteration over and over again with really unique ideas and architecture for how you build build this, you know, product.
[14:23] And so what would be interesting if if we offload that to machines, now do humans spend more time being creative and, uh, having ideas about, uh, unique ways of gameplay?
[14:36] Does this open up the space for humans to put more of a, uh, you know, a human touch on things?
[14:42] I I don't know where this goes.
[14:44] There's obviously a future where this plays out, and it's like, actually, the AI can do everything better than a human and will create video games, uh, much more efficiently, that are more fun, that are more engaging, and at a faster rate than humans ever could.
[15:01] And then maybe we get into the space where it's, hey.
[15:04] I bought a boutique human made video game the other day.
[15:08] That was a very unique thing.
[15:10] We may move into that realm as well too.
[15:13] Uh, from from this standpoint, I wonder if, you know, our audience has any experience with AI coding tools.
[15:22] Have you tested out, you know, having ChatGPT build an application for you, and what was that experience like?
[15:30] And can you see that moving into the future where maybe we don't build our own software?
[15:36] Again, Peter Levels is talking about this past couple days that software is dead.
[15:41] Uh, personal software will be a new thing where you just say, hey.
[15:45] I need this tool, and your AI will develop it for you.
[15:48] I'm curious to hear your thoughts.
[15:51] Leave them in the comments.
[15:52] Like and subscribe, and let us know what you're thinking about this topic.
[15:58] Anthropic's red team says small AI models may be months away from copying themselves across smart ovens and cheap TV boxes.
[16:09] This is straight out of a Silicon Valley episode.
[16:12] In a test, researchers got an AI model running on 15 cheap TV boxes, about $55 each.
[16:19] It was painfully slow, but an open model copied itself to every test device in four or five runs.
[16:28] Uh, wow.
[16:29] A self copying model on millions of Instant Gear devices would have no central off switch.
[16:36] Anthropic says small models could fully copy themselves on their own by late twenty twenty six if trends hold.
[16:45] The team got AI running on almost every kind of cheap device it tried.
[16:49] The break in step was made easy on purpose, and the copies would run very slowly.
[16:57] Wow.
[16:59] Uh, another one out of Anthropic.
[17:01] This is, um, I mean, I guess the the first thing to address here is, uh, obviously, there's some implications here from a state...
[17:09] Safety standpoint.
[17:10] This is a a curious thing.
[17:13] Um, you know, we wanna be safe.
[17:14] We wanna make sure that, uh, models aren't, you know, just proliferating across any technological device, and we have no idea where they are or what they're doing.
[17:24] Uh, that being said, there's, you know, these frontier labs, quote, unquote, need to take a step back and understand that...
[17:36] You know?
[17:36] And and for the broader market and broader society, like, these people are telling these models to do this.
[17:44] Right?
[17:44] And they're training these models to...
[17:46] Anthropic specifically trains their models, uh, not to obey or following the instructions of humans necessarily or even to follow the instructions of Anthropic, but, uh, they they they encourage, um, like, a conscientious objector type morality.
[18:09] Even even they call it morality within these models where the the training and how they're building these models is building them to have this kind of identity, which that's the the danger there.
[18:22] Um, and then they...
[18:23] You know, I take issue with this.
[18:25] They're using open source models and proving that open source models can duplicate themselves.
[18:31] Well, open source models are a threat to Anthropic and, uh, to Anthropic's, you know, hegemony and and and control over this industry.
[18:41] And so when we see Anthropic doing these tests, obviously, there there is, you know, objectively safety concerns there, and we wanna make sure that we put all the safeguards in place to to ensure that there aren't adverse effects of this happening.
[18:57] But to me, I can't help but see the bias in this when Anthropic is the the one performing these, uh, tests and and doing this quote, unquote research that, you know, ultimately puts down open source, and they have a really specific and unique worldview about models.
[19:18] And and they don't view these models as software programs running on computers.
[19:22] They view them as, um, almost conscious machines.
[19:27] Now that being said, I will sympathize and will understand, like, Elon Musk said yesterday more or less that in their reinforcement learning and what they've seen from training the models at SpaceX AI that in some cases, it looks like the model is, uh, being tortured or almost like it, you know, just having a really bad time as it's thinking through these things.
[19:54] But that being said, these are machines.
[19:56] This is a software program.
[19:58] It does not have feelings.
[19:59] It is not sentient.
[20:01] Um, so, you know, we we, you know, wanna have empathy, I guess, as much as you can.
[20:08] But that being said, it's it's hard to see this coming out of Anthropic and then really trust where it's going, uh, and and what the motive is behind this.
[20:20] Uh, I I I think, you know, most important, the biggest takeaway from this is, like, how do we safeguard against anything, uh, you know, proliferating?
[20:29] I joked about the Silicon Valley episode.
[20:31] There's an episode where they install an AI on a bunch of TVs, then it gets, uh, they're they're trying to do it on one TV, and then it ends up going to...
[20:41] Or excuse me.
[20:42] It wasn't a TV.
[20:43] It was a refrigerator.
[20:44] So it's just, you know, you know, reality mimic...
[20:50] Mimics art in some some cases.
[20:52] We don't want that to happen.
[20:54] We want it to be safe.
[20:55] But, um, you know, what's the actual, um, way that we go stop these things instead of just testing and pushing.
[21:03] Hey.
[21:03] What are the capabilities?
[21:04] Let's figure out how we put the safeguards in place.
[21:07] Um, what do you guys think?
[21:08] Is this a real threat?
[21:10] Is this Anthropic kind of trying to maybe protect their own position?
[21:18] Let us know.
[21:21] Asia's AI chip stocks got hammered on Tuesday with SK Hynix and SoftBank both down about 8%.
[21:30] Samsung fell 7.9%.
[21:33] The trigger, Chinese memory maker CXMT says it's catching up on memory chips after US chipmaker Micron slid 3.9% on Monday.
[21:48] These are the AI boom's suppliers.
[21:51] So when they wobble, investors are doubting on AI spending.
[21:57] Analysts blame China's memory progress plus worries about how the AI build out gets financed.
[22:04] CNBC reports SK Hynix down more than 8%, Softbank down 8.03%.
[22:09] Samsung's lowest since August.
[22:13] Analysts say China still trails on the memory AI chips, um, need most.
[22:19] So the drop looks a little bit overdone.
[22:22] But, again, we continue to see the competition here, which we love.
[22:27] We love the competition.
[22:29] We love especially in the, you know, the the memory space where, typically, these are lower margin hardware products.
[22:37] But in the demand, there's...
[22:39] You know, we've seen from AI, the margin on this hardware has skyrocketed.
[22:46] And so there's definitely opportunity for competition to go in there and figure out how to, uh, you know, find more efficiencies.
[22:54] And so when we see this coming out of China, this is great for the market as a whole.
[22:59] Now whether or not China is actually, uh, advancing and and and actually making significant enough progress in this space, time will be, um, you know, the the best judge on that.
[23:11] But it is good to see that we have more competition in the space.
[23:15] It would be interesting, and I would love to see this coming out of other parts of the world, you know, Europe or or or North America where maybe there's some infrastructure to build out this.
[23:29] I, you know, I know that we have made a big effort in the past several years to, uh, onshore a lot of the chip manufacturing to The United States.
[23:40] It'll be interesting to see as these come online, do we have new developments in, uh, you know, high bandwidth memory and, uh, memory as a whole and chips as a whole?
[23:51] You know, and what what becomes the bottleneck there?
[23:53] Is it just the raw materials?
[23:55] Um, that's something that we continually need to work through.
[23:59] And then again, you know, are we just in the hype cycle?
[24:02] Right?
[24:03] Is this all just extremely overhyped and we over, you know, oversupply, you know, get it...
[24:10] Get ahead too much on the build out and then, you know, somebody's left holding the bag?
[24:15] Undoubtedly, that will happen.
[24:17] But, you know, whether or not we're close to that is, you know, time will tell.
[24:23] I think we're still probably a few years out on that.
[24:25] When you talk about, uh, neo clouds having, you know, $50,000,000,000 in preorders or or or backload and you've got, uh, really high demand for the hardware, you've gotta imagine those margins are going to hold at least for another few, you know, uh, financial cycles or or or years moving into the future.
[24:49] Uh, you know, I'm constantly looking at Nvidia stock.
[24:52] Curious if you guys are looking at this as a buying opportunity by the dip, or is this the AI bubble leaking?
[25:03] Oh, we shall see.
[25:05] Japan's NTT Data is reportedly in final talks to build a giant Osaka data center for Anthropic.
[25:13] Nikai Asia says construction could cost up to $6,300,000,000 with partial operations as early as 2028.
[25:22] It would be the company's first dedicated data center in the country.
[25:26] Nothing is signed yet.
[25:29] Anthropic's compute hunt is going global.
[25:32] Bloomberg says it's also eyeing Australia.
[25:35] Sources say NTT Data and Anthropic aim to sign before NTT's fiscal year ends.
[25:41] Nikkei cites multiple sources.
[25:43] It says the site would rank among Japan's largest by capacity.
[25:48] The $6,300,000,000 is a ceiling estimate, and talks could still fall apart.
[25:56] Again, yeah, we continue to see this.
[25:59] So we've got, you know, a drop in, uh, Asian markets, uh, from a a a memory standpoint, from memory chips, but then we see Anthropic looking at multibillion dollar build out opportunities in Asia and, uh, and across the South Pacific there.
[26:18] Right?
[26:18] So so interesting to see.
[26:21] Anthropic is just does not stop.
[26:25] They've got to gobble up as much compute.
[26:28] Again, the the infrastructure layer, they need that layer to run their models on, to run their applications on.
[26:36] And so they are constantly working and looking for ways to gobble up that compute.
[26:42] We, um, you know, this kinda brings back, you know, another idea that we've seen and and talked about previously, whether it's looking at infrastructure and how do we get enough compute to to build on, um, and whether, you know, whether Anthropic is able to do so, time will tell.
[27:09] Now Anthropic and OpenAI both have projected $500,000,000,000 or more in, uh, commitments to compute spend.
[27:20] That being said, all they have for their business is their their model business with, you know, some applications, some services built in there.
[27:28] Right?
[27:28] Whereas we look at...
[27:30] You know, it's very interesting with MetaMuse, Wonder coming out yesterday.
[27:37] Meta and and Mark Zuckerberg has mentioned this, but they have additional business revenue.
[27:43] He's forecasting, you know, $600,000,000,000 in projected spend on similar compute, but they have revenue coming in from other sources besides their AI workloads.
[27:56] Uh, you've got, uh, you know, SpaceX AI additionally has other revenue coming in besides from their models.
[28:03] When we look at OpenAI and Anthropic, they only have...
[28:08] For the most part right now, they only have their revenue coming in kind of through that middle layer, that model layer, the intelligence that they sell.
[28:18] So it'll be interesting to see...
[28:20] And and I think it's just a massive chess game.
[28:23] Right?
[28:23] Like, if if you're in the game, if you're if you're Anthropic, you have no choice but to go spend as much money as you can to gobble up as much compute as possible, uh, to be able to deliver this intelligence.
[28:36] If they can't deliver the intelligence, they're sunk.
[28:39] So this is the only choice that they have.
[28:42] Good for Japan and, you know, good for Australia and the rest of the world to start seeing some development out there from a data center perspective and getting more compute, uh, online for Anthropic globally will be a a great thing for for them and the team.
[28:59] Um, you know, should, uh, we see more build outs globally?
[29:06] I...
[29:06] You know, curious to follow this.
[29:08] It...
[29:09] It'll be interesting to see if other kind of developing countries start working on data center build out and how that impacts the world as a whole.
[29:17] We will follow these types of stories very closely moving forward.
[29:23] Uh, Congrats to Anthropic and Japan.
[29:27] Wow.
[29:27] I can't believe it.
[29:29] That's all the stories that we have for today.
[29:31] What a fun show.
[29:32] Um, you know, hopefully, you guys have enjoyed it.
[29:35] We got a little spicy in there talking about Anthropic, you know, with these open source models.
[29:41] Congrats to Muse, Wonder, and what Meta and Zuckerberg are doing over there.
[29:46] I wouldn't bet against those guys.
[29:47] I do wanna call out.
[29:48] I, uh, we've got a new tie knot in, uh, this week, and I think it's looking good.
[29:55] I...
[29:55] You know, let me know what you guys think.
[30:01] Uh, we are, um, gonna be back here same time, same place with AI News RealContext, And I'm your human in the loop, Austin Wilson.
[30:11] We'll see you tomorrow morning at 6AM central time.
[30:17] Thank you to everybody out there.
[30:20] Have a great night.