AI News Weekly ๐ Copilot Edge, Open-Source Model Wars, Tencent 3D, & Trump-China Regulation Clash
Hi, everyone. Welcome to the monkey patching podcast, where we go bananas about all things bubbles, browsers again, and more. My name is Murilo. I'm joined as my side host, Bart.
Bart:Hi, Murilo.
Murilo:How are doing, Bart?
Bart:I'm doing good. I'm doing good. I have a bit of atypical surroundings as you can see behind me.
Murilo:Indeed. Really cool.
Bart:Really cool. Actually on vacation. Wow. So it's a bit of a This is the vacation. Holiday sessions of the monkey patching podcast.
Murilo:Indeed. We still the listeners will see us a bit different, but we'll still have
Bart:stuff to talk about. We'll still try to give a weekly update. It'll be a bit shorter than you're used to, and it depends on me having a connection.
Murilo:No. But we all appreciate you making the time. It looks like it looks like you're a very cool spot, so definitely enjoy. But then let's get to it. Let's get to it.
Murilo:What do we have, Bart, starting up?
Bart:We have a technology writer at Zetron tears apart what he calls the overhyped generative AI gold rush, warning it's prompted by wishful thinking and unsustainable spending. As he bluntly puts it, we're in a goddamn bubble, prompting a no holds barred discussion on whether the boom will burst. So he has this very cool cool titled article called the Hetero's Guide to the AI Bubble, which is, of course, a play on the hitchhiker's guide to the galaxy. Were you able to read it, Mariela? What do you think?
Murilo:I had a I had a look into it. I studied a bit. It is a it is a long article. I think even he mentions, like, one last thing yeah. One last thing, this new letter is nearing 14,500 words is long, perhaps consider making a pot of coffee.
Murilo:So I think he's also well humored as well. But the author really goes over basically everything, not everything that is wrong, but why is it so much a hype and he defends a bit why this is something that we should be mindful of. So a few things that he mentions here is like NVIDIA's dominance on the market. So we have the other companies there that also profit from the AI thing, but most of them actually rely on Nvidia and he actually made some points like Nvidia's growth relies on them selling more chips every year, which is just they're assuming that they would do. He also mentions how most companies actually are not making a lot of profit from JNI.
Murilo:It's basically a lot of hype and talk, but it's not really revenue yet. And I think again, goes to people invest, even though people are investing on this a lot. They also mentioned that there's a lot of there's no value proposition, quote unquote, right? In the sense that there's a lot of people doing a lot of stuff, but there's a lot of overlap. A lot of times it's very easy to kill a lot of these startups that they're trying to show new value because really like and we're going to see as well like browsers as well in the next topics.
Murilo:Right? So it's not sustainable as well. He also doesn't like the terminology, the fact that people say agents for everything these days. So he kind of gathers a whole bunch of arguments to say that this is a bubble, which the way I read it is a bit like this is hype. Right?
Murilo:Don't get too fooled into it and why he doesn't he's not a big he's not riding the train, let's say. He's not on board of the train the hype train.
Bart:Not riding
Murilo:the hype train. Yeah.
Bart:He's not riding the train. In community, there are definitely people that are pessimistic about about what AI is, what it could do, or the state that it's in. I think there are also a lot of haters, which you probably have to whenever any big evolution occurs. And I think if you if you are a hater, I think this is your perfect bible. Because the
Murilo:Yeah. Sure.
Bart:Arguments he lays out, like, I mean, they are you can't really have anything against it. Like, he's saying, like, there's a huge ecological cost to these things. The capital expenditure is way, way higher than any revenue that's being generated on this today. And it's very much like going a bit against the extremes and also a bit against this sort of a gaslighting optimism where everybody's saying, ah, we're doing this. We're building AI so that we can decrease our workforce.
Bart:So these are very weird things to rally behind, which we do see every day in practice today. Right? That these things are being being positioned as very positive, while a few years ago, we would have very much questioned this. So I think I understand where he's coming from, and I do agree with definitely with some points, but I I don't really follow the the the part where the value in the future is not there because we don't see the profit today. And getting to that value, I mean, will probably go at huge ecological cost.
Bart:But Yeah. That's the difficult thing of this.
Murilo:Yeah. Indeed. Indeed. I think, as you mentioned, there is a it's very it's very appealing for people that want to hate on the train, on the hype train. I think like you said, think it puts arguments as well that maybe also wonders like, oh, yeah, he has a good point.
Murilo:But I think in its totality, I think it's a bit I also wouldn't go as far, right, to say like it's a bubble, that it's hype and it's this and it's that. I I think there's more to
Bart:really Of course, to some extent, it is hype there, and, like, it's being positioned as the answer to everything. And it's definitely not. Like, if you see where it is today, like, it's very good at coding, but it will not brew you your next cup of coffee, right, and bring it to you.
Murilo:Yeah. Indeed. Indeed. I think, actually, I
Bart:also think today and that there's promise in other areas.
Murilo:Yeah. I also think that, like, that's a bit the problem. Like, it promised so much, the whole Gen AI, Agent Tech AI, that even if it does a lot but it doesn't do everything that it promised, some people see it as a failure, which is not true, right? Like it still does quite a lot of stuff. So yeah, indeed.
Murilo:And I agree with you that there are also some other points to be considered, right? Like the environmental impact, there's a lot of think OpenAI had the Stargate project, right? There's also stuff from China, there's a lot of stuff as well that shouldn't be overlooked, right, which I think brings maybe brings us a bit to our second topic, and I'll skip a bit around the the order. That Washington's latest AI blueprint combines export zeal with lighter green rules aiming to outpace China by shipping full AI stacks, quote unquote, to friendly nations. By kicking off the push, Trump proclaimed, and I quote, America is the country that started the AI race, end quote, A line sure to spark global tech power chatter.
Murilo:Again, a bit related to what we were saying, think Trump administration, they unveiled a bit these what is it? Is it a
Bart:It's AI plan.
Murilo:It's a An AI plan. Yeah.
Bart:It's a it's a big plan. It's to become a bit like the the or to become or to stay the dominant global player in AI. It's a long plan. It's 90 points. I've I've went through it, tried to understand a little bit.
Bart:All 90. There's also no. Not all 90. There's a bit of vague terminology, so so I'll I'll try to explain a little bit. What they want to do is that they want to form AI alliances with friendly partners across the globe.
Bart:And in those AI alliances, they want to have no restrictions on export and import On, for example, AI chips, so hardware, software, services, whatever, to really strengthen their position in the AI game. The important nuance here is that those AI alliances, because the question there is like, who will this be? And the vagueness that is being introduced is that what they're saying at the same time is that they are scrapping the Biden area restriction on exports of AI chips, which sounds like now the exports to China will start up again. But it's actually not like that. That is not made concrete because it's not clear how these alliances will be formed, with which partners, how the selection will be done.
Bart:It's clearly unclear, like, what the definition of an AI alliance is between countries or regions. Next to those alliances, you also like, there's also a focus on speeding up data center production across The US To speed up is also they're scrapping some ecological barriers. So certain limitations that they had to adhere to, they don't no longer. So also raises questions around sustainability, of course. Another important thing is that that this is at the federal level, so that means that it will basically Consolidate the regulation of this of all all states in The US.
Bart:So they will have one policy on this going forward. But let's see. Like, it's it's a it's a big there's still a lot of vagueness, especially on what are what is the definition of AI lines and how would it shape up?
Murilo:Yeah. I think I also I also came across other articles that they were talking about China closing the gap in the chips race or the AI chips race and all these things, which to me feels a bit like it's linked to that. I think it's
Bart:a fear to lose a bit this position as the leader in AI today.
Murilo:For sure. I also wonder if this has an impact on the EU, in the sense that EU is falling behind a bit when it comes to the AI race.
Bart:Question is, have the EU has ever been relevant? Aside from ASML? Yeah. And, actually, I think it's it's not officially linked. They didn't mention it, but the timing is very interesting is that I think yesterday at the Global AI Conference, China, basically, premier of China, they proposed a global south organization to basically strengthen collaborations on across the global south on AI.
Bart:And it seems a bit as a reaction to this this plan that is being proposed by by Trump's government.
Murilo:Yeah. We'll see. We'll see. I think also it's a bit like you mentioned, like you being relevant. But I think things like this to kind of scrape the whole environmental aspect to just say, let's just go fast and consolidate this position.
Murilo:It's it's also something that wouldn't happen to you probably. Right? Like, it's it's a bit
Bart:Yeah. I find it's, like, also that you have these two very big powers. I find it interesting that there's just a quote unquote technological evolution, which is AI, and you have these geopolitical movements around us that almost feels like an arms race.
Murilo:Yeah. Indeed. Indeed. Yeah. It gives me vibe when I was in high school in Brazil and you're studying, like, the the cold war, you know, and
Bart:the Yeah. Exactly.
Murilo:Race and space and all these things. That's that's the vibe that I get.
Bart:Yeah. Yeah. The race to it. Yeah.
Murilo:Yeah. Yeah. The race to the end. Yeah. Yeah.
Murilo:Yeah. Cool. What else do we have, Bart?
Bart:We have Microsoft. Microsoft has a new Copilot mode in Edge, and it turns Edge Edge into their browser. It turns Edge into a chat powered helper that can juggle tabs, organize research, and even take voice commands. The update greets users with a single input box combining chat, search, and web navigation features, a neat cue for how browsing may soon feel. So, yeah, this is, I guess, to be expected.
Bart:Right? We've been talk talking about the AI browser a lot. Microsoft has their own Yeah. Browser. Microsoft is very big in the AI game, especially with their OpenAI partnership.
Bart:So now they bring Copilot to Edge. I haven't tried it.
Murilo:I'm not sure if it's
Bart:all available here already. Not sure if there's any what the pricing will be on that. Looks interesting. And then they're they're positioning it really as a as a single input. So, like, with with that same window, you can do search.
Bart:You can do you can ask questions. You can drill down. You can interact with different tabs, etcetera.
Murilo:Yeah. I think we talked about a a future like this
Bart:Yeah. Yeah.
Murilo:Previous episode. Right? Like, I think they also even mentioned that they want to
Bart:Well, indeed. Yeah. We discussed another brow like, there are a lot of these browser projects today. Right? But we didn't really discuss Microsoft, which is, of course, the elephant in the room because they're big in AI and have a browser.
Murilo:Yeah, indeed. Indeed. They even mentioned that in the future they want to add additional context like history and credentials so people can actually start booking stuff and stuff like that. Right? So again, I also feel like this is killing a whole bunch of projects and startups as well.
Murilo:Right? Like it's one of those things.
Bart:Exactly. Yeah. Yeah, exactly. Especially from Copilot and Edge browser can also like take actions, right? Like if it can actually browse your web pages, if if it could do click commands, if it can fill stuff in, like, it goes very quickly then.
Bart:Right?
Murilo:Indeed. Indeed. But I'm curious because I think you tried one browser before that had a bit of that promise, but you said it was really bad.
Bart:That's a while ago. Yeah. That's a while ago. But I actually tested a bit more with ChatGPT Agent that was recently released. They they use a they call it a remote browser, so it's a bit of a browser instance, but you connect to it.
Bart:So that also means that, like, you can close down your window and it will still complete in the background. And how they did it is and I actually found an interesting premise is that, like, I can ask, find me that flight and book it for me. But then I can also, like, for for example, from the moment that I need to fill in my credit card credentials, I can take over the browser and I can fill it in myself.
Murilo:I see. I see.
Bart:Like you can do these things and that's, and it feels I actually played around with it a bit today and it feels quite fluid. For the way that the way that they positioned it. And I think if you're in your actual browser, which you will be with Edge, like, can make it even more fluent.
Murilo:True. But at the same time, you're also constrained to the browser world. Right? And maybe maybe even if you're if you're on the agent and the experience already feels fluent, maybe there is an argument to just be broader than that. Curious.
Murilo:Like, maybe we have to to deep dive one day on this OpenAI agent thing. Very
Bart:curious. Think it's gonna be cool to do a session on us. But I think they're also, like like, the I I think three years from now when you see someone, like, calling a restaurant to make a reservation, you you will think the person is completely outdated. Right? Yeah.
Bart:Like, this is this old dealer that doesn't know how his agent works.
Murilo:Yeah. I mean, to be honest, even today to a certain extent, there's a bit already, like, why are you Yeah. You know, why are you calling? Like, you can do async. It's like
Bart:That's true.
Murilo:I don't know. But it's true. It's true.
Bart:Actually, the trip that I'm now on, I I actually did a lot of selection of spots using ChatGPT Because, like Really? We visit a lot of campsites, but we want a bit, like we prefer some scenery. We prefer some certain activities around it. We prefer some, like, maybe close to the seaside. Maybe they maybe maybe an alpine terrain.
Bart:And, like, there are so many options. Like, it takes so much time to visit all those websites. Like, it's super easy when you do, like, a deep research with ChatGPT to do these things. Yeah. So I
Murilo:think for me sometimes, it's a win just because there's so many options. And after a while I kind of don't care, like as long as it's one of those three, I don't care, but I think just having someone say okay just pick this one, was like yeah okay let's go with this.
Bart:Especially when you can define like these are my requirements. Right? And then give me the 10 best options, and then I can fine tune.
Murilo:Exactly. Yeah. Exactly. So, yeah, I think it's also for me, it's a it's a life improvement in that sense. Because I can even though I'm getting bad at it, like, having the choice paralysis, like, think I it helps it helps with these things as well.
Murilo:And you live more. AI. Yeah. Helping you live more. There's a there's a logo for you.
Murilo:What else do we have? Should we go to the next topic? Yes. Alright. So we have Beijing based Xipu.
Murilo:I'm going to guess that's how it's called it. Z H I P U has open source GLM 4.5, pitching it as fresh fuel for intelligent agents and adding yet another model to China's bulging roster. The company says the new release is, and I quote, designed for intelligent agent applications, a phrase hinting at growing ambitions beyond playing chatbots. I actually didn't know about this company, but apparently from the article, even OpenAI acknowledged this startup, or maybe I didn't know. I just didn't know the name.
Murilo:Have you tried this model actually, Bart? GLM 4.5?
Bart:No. I haven't tried it. It was introduced it was released, I think, yesterday or the day before. It was very recent.
Murilo:Yeah. The article is from yesterday. So
Bart:Oh, yesterday. When you it's a it's a mixture of exports model. It's the total all exports together contain 355,000,000,000 total parameters. Crazy big to think
Murilo:about it. Yeah.
Bart:It is today the best performing open source model. It outperforms Kini that we discussed, I think, two episodes earlier. It also outperforms DeepSeek.
Murilo:And Lama is old news at this point.
Bart:Lama's old news at this point. Exactly. And it's it's got a lot of funding also. The Zipoo is a relatively new company. It's also apparently very heavily local government backed.
Bart:What was interesting to me, and I I didn't notice, but, like, it's the amount speaks volumes. Like, it's China's 1,005 hundred and ninth LLM, which is Yeah. A lot. Right?
Murilo:Crazy. Yeah? It's crazy. I think it also links a bit to the the who versus China race story. Right?
Murilo:Like, you do see China. There's a lot of stuff they're doing. Yeah. This is One thing one thing as well.
Bart:This the pace is like, this completely outpaces Europe. Right?
Murilo:Yeah. One day I thought it'd fun
Bart:to we get a new a new update about the new Mistral model.
Murilo:Hey. Yeah. That's true. And it's gonna be, like, disappointing. Or it's gonna be a bit like I
Bart:mean, it's
Murilo:gonna be really good, but these things are going so fast that it's
Bart:like Yeah. Yeah.
Murilo:If it's not the best, it's a bit like, yeah. I'll wait for something One thing I thought it was funny, at first thing I saw is a GLM 4.5. It does if you squint your eyes,
Bart:you do GPT
Murilo:4.5. I thought it was funny. I thought it was funny. But it's cool. Again, very curious to see how these models as well.
Murilo:I also heard that Gwen, I think Gwen also had an open source model, they also had the Gwen CLI, and I think it was one of the first ones that seemed to be comparable to Anthropic for coding. I think that's something that excites me a bit, kind of seeing how people are doing, making it open, you know, trying to level a bit the playing field.
Bart:I think it still costs a lot to train these models, but apparently, a lot of companies are able to do it. And I think that's good news, right, for global competitiveness. Interesting is that I was looking at the chatter on Reddit. Some people try to just have a simple hello world, like, hello, who are you? And, like like, half of the time it answers, I'm zPoost GLM 4.5.
Bart:The other time it says, I'm this model from Anthropic. Really? Yeah. Yeah. So but people don't it's the it's their hosted it's it's their hosted API, And people don't really know if it's because it's trained on training data based maybe on Sonnet, which is done a lot.
Bart:Right? And, like, a a lot of these companies generate training data by using other models. Or maybe just because when their API is is called too much, like, there's too too heavy load on the service, then maybe it may be proxies to Anthropic to still be able to to actually give the user a response potentially.
Murilo:I see. I see. I see. I see.
Bart:That's the positive. But it's also
Murilo:yeah. Yeah. Yeah. But if it is the latter, I think it's also interesting that they can they can get away by being distinguishable with Entropic. Right?
Bart:Yep. It's actually it's a the hosted model is very cheap. Like, makes if you if you just look at money versus quality, like, it makes a lot of sense to use it. If you look at the benchmarks, I haven't tried the model myself yet.
Murilo:And then if you wanna host these things easily, can you use OpenRouter for this or no? Not you host it up, but pay for
Bart:these Yeah. Think they're available through OpenRouter, yes.
Murilo:Very cool. Anything else you want say on this part? I think those are the topics that we wanted to dive deeper, but we have more more stuff, right, that we wanted to just share, and we put the links on the show notes if you would like to have a look. Yep. What do we what else do we have, Bart?
Bart:We'll read them one by one. Normally, we do a discussion on every article, but because we're in in a holiday mode, we'll just read the the headlines. So we have the Hunyan. Not sure I'm pronouncing this correctly. Hunyan.
Bart:World Model one point zero. Tencent's Hunyan team just dropped its first open source engine for instant explorable three d worlds, stirring excitement across game and VR circles. Their launch tweet beams were thrilled to release an open source Hunyan three d world model of the one point zero, inviting hosts to imagine the possibilities. Very cool model, which is a quick note on this. I looked into it.
Bart:It's you use a few text prompts, and you basically through a few steps, the paper explains it quite neatly, you generate a three d world, and they also provide you with a HTML based world explorer. So you can load that three d world into your browser, you can actually walk through it. Very cool. It's actually cool. But if you go to the website, there's a lot of videos to give you a bit of a feel of what you can do with it.
Murilo:Very cool indeed. Also, it would be fun to see some projects on this. What else do we have? We have OPAL. Google Labs has unwrapped OPAL, a drag and drop playground where non coders can chain models and prompts into bite sized AI apps.
Murilo:The post kicks off with, and I quote, We're excited to announce Opal, a line that invites makers everywhere to start remixing AI workflows. So this looks like there are a lot of low code, no code solutions, right, for the AI apps, the Gen AI apps. N8N is the one notable one there, which actually I heard the other day. It's supposed to be spelled Nathan, which I thought was clever, but maybe too clever. Wow.
Murilo:I didn't
Bart:know that actually.
Murilo:Yeah. That's what someone said. But so I guess this is a solution from Google, drag and drop things. So if you're not as techy but you would like to bring your AI app to life, this is the place or this is a place. Right?
Murilo:What else?
Bart:Then we have a developer, Will McGugan, known from Textual, a Python TUI
Murilo:And Rich.
Bart:Library. And Rich. He has a prototype to TOAD, a flicker free terminal UI meant to tame and turbocharge agentic coding workflows. He quips, I'm a little salty that neither Anthropic nor Google reached out to me before they released their AI coding agents. Throwing playful shade.
Bart:You're a McGoogan fan. Right, Marillo?
Murilo:Well, a fan is a strong word, but I think I've used some stuff that he built. I think it was pretty cool. I think this one so he talks about building CLI for agents. He actually talks a bit even about architecture and how you can do things. He also criticizes a bit the the cloud code and the Gemini CLI.
Murilo:It's written in JavaScript. So I think it was I thought it was also interesting to see here a bit of his thoughts on what could be done. And this is not a project that is open source yet, but he probably will open source in the future. He's taking a sabbatical from Textrol, the company that he founded, but he also sunsetted, let's say, he stopped because he wasn't working out. So I'm curious
Bart:I'm a little bit skeptical on this.
Murilo:Yeah. Did you read this?
Bart:I read yeah. I went through it. I'm skeptical because, like, he's positioning Thoth like this is new CLI based agentic coding thing. But the main premise he's making on why he's creating it is because there is flicker in the Yeah. Cloth coats and the Gemini CLI, which I never noticed.
Bart:I mean, that should not be your main promise. Right?
Murilo:Yeah. He he he's a you can tell he's a CLI geek. Right? I mean, probably would compare it to this. To the two users.
Murilo:Indeed. Indeed. So I think, like, when he sees something
Bart:like this competition is all competition is welcome on this, and there are actually a number of open source projects. I think we discussed a few in the past. But the promise of a CLI should be more than there is zero flicker.
Murilo:I think so too. I think so too. I mean, he does mention as well, like, architecturing the applications, how we can grow, but I think it's True. Yeah. I think it could be nice.
Murilo:I think I'm curious to to see exactly what it is. And my what could get me excited is me building, like, little CLI applications with AI, things to try to help my life. But let's see. Finally, we have Alibaba. Actually, this is the coder.
Murilo:Alibaba's new GWENT three coder model enters the cogeneration fray, claiming boosts that rival both domestic peers and Western heavyweights. A company pitches it as, and I quote, its most advanced coding tool to date, end quote, a boast that rises the stakes in China's escalating model wars. I think actually I alluded to this and actually I forgot that we had it at the Hey, I mentioned
Bart:it earlier, right?
Murilo:Yeah. I preluded this, let's say. So GWEN three, I think it's an open source model, I think.
Bart:Indeed.
Murilo:And it's from Alibaba. And they also created a CLI tool to rival Cloud Code. And it's actually a fork of Gemini CLI. So we were talking about CLI That's coding what I heard as well. So Gemini CLI is open source project.
Murilo:So they actually forked it and created Quent three. And one thing I saw is that it's cheaper to run. I think it has less parameters, but I think it's cheaper to run open source. And it was one of the first ones that I saw that rivaled Anthropic in benchmarks on how good of a coding tool it is. So haven't tried it.
Bart:Yeah. I'm very much niched down on being a coding assistant. Right?
Murilo:Yeah. Yeah. Yeah. Yeah. I think so.
Murilo:I think so.
Bart:Think more specifically on multistep agentic programming tasks, which you typically see, like, that process you you see when you use something like Gemini or a or a ClothCode where you have multisteps to come to a certain results. That's apparently, that's where it's really trained for.
Murilo:Yeah. Indeed. Indeed. Which I mean, if it works well, fits your need, you know what to do, you know who to go to when you have a question, I think that's also fine.
Bart:It's it's weird to that in today's world, like, an announcement like this two years ago, a new Gwen model would have been major. And now it's already surpassed by what was it called? The Zephus model.
Murilo:Yeah. The g g l m 4.5.
Bart:So this was released Alibaba's was released on the July 23, and then Zippus was the twenty eighth. So, yeah, they had their time to shine. All the
Murilo:Tomorrow, time
Bart:we'll have
Murilo:tomorrow, we'll have another one. You know? So it's like it's just like that. That's that's that's the world now. But that's the thing.
Murilo:It goes back a bit to what we're saying. Yeah? It feels a bit like of a AI race. Right? Yeah.
Murilo:True. It doesn't like, I don't know. Yeah. But that's it. That's what we had for this this week.
Murilo:Anything else you wanna add, Bart?
Bart:No. I think it was fun to do the short session. And I think what we should also do, like, discussing a bit like this, there's this model, there's that new model, but I think maybe we should also dive a bit deeper into how can you use this? What are the use cases? What are some success cases in the industry?
Bart:I think that would be interesting as well to search that out a bit more.
Murilo:I think it could also be nice to have a bit of a, like you mentioned, the OpenAI agent to kind
Bart:of share a bit
Murilo:like, oh, this is what we did. This is really cool. These are first impressions. So I think that could be really nice as well. But another time in another place, not on a quick, quick round mode, maybe.
Murilo:But in any case, Bart, thanks a lot for joining. Enjoy your time there. Enjoy the sunset.
Bart:I will. I will. A big thank you to our listeners. We love all the interactions and feedback. And if you have any thoughts you want to share, head us up wherever.
Bart:Indeed. Don't forget to subscribe.
Murilo:Exactly. And also, if you like this format a bit shorter and just like news at the end without too much deep dive deep dives, let us know as well. You know? Always happy to to hear your feedback. Alright.
Murilo:Thank you. See you next time.
Bart:See you next time. Ciao.
Murilo:Ciao.
Creators and Guests
