Copilot Slowdown, AI Taxes, OpenAI Images, ty/Toad, ALPR Watch, and the 300TB Spotify Mirror
Hi,
Murilo:everyone. Welcome to the monkey patching podcast, where we go bananas about toads, Spotify, and more. My name is Murillo. I'm joined by my friend Bart. Hey, Bart.
Bart:Hey, Murillo.
Murilo:I slipped a bit there. What did you do? I slipped a bit. I don't know. Was like, can we do this every week?
Murilo:But still sometimes I'm like reading the words. I trip on the words or something.
Bart:That's okay. That's okay. It's good to see you in person.
Murilo:Yeah. Indeed. But last time we saw each other in person was what? Two weeks ago?
Bart:Yeah. That's true. That's true. But normally we we had interview with Charlotte. Normally, we we record remotes these days, but we're going for us for dinner.
Murilo:Exactly. Maybe also you mentioned the interview with Charlotte was really cool. Actually, we have quite a lot of views so far, but like very, very nice interview about the the how funders fit in the Belgium and Belgium fits in the EU strategy for AI, the UEI act. We talked a lot about stuff. So another open invite for
Bart:Yeah. It was very interesting. I learned actually quite a few things there.
Murilo:Yeah. Yeah. Again, like, she's super knowledgeable, super casual talking about stuff, but, indeed also was very nice to see the, the other side of how these policies get made. Exactly. So it's very, very cool.
Murilo:Maybe before, just before we start as well, you know, thing that I learned the other day, I mean, and I didn't know this actually, and I felt a bit silly for not knowing.
Bart:Curious what's coming now.
Murilo:That the pod in podcasts, do you know why it's called podcasts?
Bart:And that they don't need to make me look bad.
Murilo:So, cause I was like, am I cause actually it's from iPod. Like, Apple started this thing, and that's why it's called podcasts. That's that's what I heard. I heard it on another podcast, but I didn't check. I feel like I hear these things that I just propagated, but, like, I maybe I should check before.
Bart:I I That could make sense. Right? Like, it sounds logical.
Murilo:Yeah. It does. Right? I mean, otherwise, what's just a pod? I mean, I don't know.
Murilo:But I I heard that. Was like, that's interesting. I didn't know that.
Bart:That's cool. At least we know what we're doing now. Right?
Murilo:At least now, yeah, we're basically promoting Apple products. Old dead Apple products. Right? Because there are more iPods anymore. It was
Bart:good, though, the iPod.
Murilo:It was good. It was good. Yeah. I feel like Apple hasn't really broken the ceiling since the iPhone. Right?
Bart:It's a whole another story. Let's get into our topic. Don't have
Murilo:time for that today. What is what do we have for today, Bart? What is the first topic that we have?
Bart:Our first topic is about Microsoft. A Reuters piece says Microsoft denied a report that lowered sales growth targets for AI tools like Copilot Studio. Even as some customers resist new agent building products, shares dipped nearly 3% before paring losses, and the company emphasized quotas remain intact while adoption and ROI pressures intensify.
Murilo:Yes. So I didn't, I don't know if I missed it, but I, because basically Microsoft is denying this report. I didn't even catch the report that they were lowering AI software sales.
Bart:Yeah. I heard about I You did? I read about the rumors. Yeah. And the rumors trigger the but it's very like, it's it's a bit of a blip on the radar, but the rumors triggered, like, a 3% dip of the stock price, later recovered.
Bart:So it was, like, it was a blip blip on the radar that was gonna be gone. Like and also, like, to maybe to, expound on the rumors a little bit. Like, the rumors are that Azure salespeople.
Murilo:Mhmm.
Bart:It's not verified, but that they are expected to drive 50% growth in Foundry related, customer spent revenue in their batch.
Murilo:Like in the region or something. Yeah.
Bart:In their group of customers, basically. How much? What's the percentage? 50% growth.
Murilo:Considerable growth. Okay. And so when they say AI software sales, it's mainly Foundry?
Bart:Foundry related. Okay.
Murilo:But it's like Azure. It's not like Copilot.
Bart:I think Copilot is actually under there. But not Copilot. It's under Posture.
Murilo:Like a three sixty five Copilot, I mean.
Bart:Maybe don't quote me on this. Okay. I'm not something sure. But in the article, actually explicitly mentions Copilot sales as well, so I'm not sure how how that translates to their targets. Yeah.
Murilo:Because I did see I I didn't get to read it up.
Bart:And it's copilot studio, by the way. Maybe that's where you are. The part of studios where, like, you have the drag and drop interface to Yeah. Quickly build your own agent that you then can connect with Copilot three sixty five.
Murilo:Yeah. I feel like they're they need to come up with more creative names. Because there's GitHub Copilot, and there's three sixty five, the Microsoft three sixty five Copilot. So it's like the the thing that can generate emails and stuff that is just there. And then there's this Copilot studio.
Murilo:And, actually, even within Copilot studio, there are different modes. There's one that you can add tools, and there's one that you can do this. But, basically, it's like create custom prompts, and then you can do certain things for you. Like, can access your SharePoint files and all these different things.
Bart:It's confusing. Yeah.
Murilo:Yeah. It is. It's very confusing.
Bart:But I think the thing why it's why it's you sought, like, the effect on the stock exchanges, these these sales or basically the sales targets in the end, what everybody assumes is that based on these sales targets, they make their sales forecasts that they also communicate, and that gives you either a good or a negative outlook on stock. And if basically they can't make the forecast that we're already seeing now, then this is Yeah. The results are coming are not great.
Murilo:So, basically, I mean, this I understand that it's a it's a bad signal because if they're lowering it, it's because people are not selling it, which would reflect poorly in the company. Yeah. And if the company is not doing as well as you think, you would sell the that's prob I mean, I'm making a lot of hypotheses there, but that's the most likely cause of why That's
Bart:basics of the dynamic. Yeah. Yeah. Exactly. And and you do you use Copilot a bit?
Murilo:Which Copilot?
Bart:Let's say Copilot three sixty five. That's easiest. No. I mean And why not?
Murilo:Well, so I don't have the the the license that has embedded in all this stuff. Okay. So usually, like, I could use it for, like, a chat GPT for work. Right? But it's not as good.
Bart:Exactly the same experience.
Murilo:Yeah. And and I I I mean, I talk to a lot of people, and I haven't heard a single person that said, no. It's pretty good. Like, it does it does everything I
Bart:To me, the only people that say, no. It's pretty good are people that do not have a do not have any experience with JetGPT or Gemini or that never really explored these things and just from their from their employer got a license to use Copilot. Yeah. And it's their first entry to the to the LLM world.
Murilo:Yeah. I think the only thing that I I can believe that it's good is, like, taking notes. Yeah. Yeah. You know, like, because it's like Microsoft, you're like, press the button, it's there, like
Bart:It's in a very connected ecosystem. It should be connected.
Murilo:Yeah. I think that the connectivity gives points to that, but as an AI model, there are better ones, I think.
Bart:Well, and they don't even have an AI model.
Murilo:Right? Yeah. That's true. That's true.
Bart:They're
Murilo:Yeah. But it's
Bart:more or less calling OpenAI, I guess.
Murilo:Yeah. But probably like an old OpenAI model. Right?
Bart:Well, they're not explicit about it. So they're probably trying to optimize cost.
Murilo:Yeah. It's true. It's true. But I, because I did remember, like, seeing article titles some somewhere going by that people from no one's using Copilot basically. So, yeah, maybe just one last thing here, like, I mean, they put this graph here, right?
Murilo:Showing the stock basically and the growth, I guess. And they were showing that Google is is ahead. Maybe it's misleading, right, because the article is about AI usage, but this is just growth in the stock. Right?
Bart:This is this is what we're seeing here is is growth in the stock this year.
Murilo:This year.
Bart:Yeah. Yeah. It's hard to say something. Think like even if let's assume Copilot isn't the greatest tool. If we assume that I think Microsoft still has a very strong position in terms of the compute power that they can give to the whole ecosystem.
Bart:They are very, very strong there. They have a very strong European presence as well, which might become strategically important. I think Alphabet here is off the charts because they do a lot on AI. I think a lot of that is, like, is valid. They make the TPUs, so they become quite good.
Bart:They make out they're making advancements in that. I think the whole Gemini three thing is way overrated.
Murilo:Yeah. I do think the Gemini three is is is overrated. And also, like, the Gemini on the Google Workspace is also it's also hit or miss. Right? It's not like
Bart:It's hit or miss. Yeah. You know? I I think since since Gemini three, everybody's like, now Google has won the AI race, but, like, if you're realistic about what came out, like, the only thing that's really better than although is that you have an image generator that is good.
Murilo:Yeah. That's true.
Bart:And that's the only thing that's really way better than the others.
Murilo:Yeah. That is like objectively. Everyone's gonna say this is yeah. That's true. That's true.
Murilo:That's true. Yeah. I think well, to be said, I think still maybe still a bit early. Right? Actually, if you think about it as well, we talked about the Linux foundation, I think last week, I don't think Google donated everything, anything.
Murilo:Well, not covering the article at least.
Bart:That's really not at this stage. Yeah.
Murilo:Yeah. I know that they came up with like a protocol before, but yeah. So, yeah, I think maybe it is still a bit early. Right. And I think maybe this growth is not there can be many things that drive this.
Murilo:Right. So to be seen. And next we have yeah. One second. El Pais revisits the quote unquote robot tax debate as AI investments source and layoffs mount, asking whether automation should help reduce lost labor taxes.
Murilo:Economists warn design is tricky. One expert notes labor provides about 85% of US federal revenue while the IMF favors adjusting capital taxes instead. So this is basically like the big discussion. Right? But the the topic of discussion is kinda straightforward.
Murilo:Right? Like, the premise is AI replaces people that do jobs, and maybe we should tax AI usage.
Bart:Yeah. And leaning up to this, we already had people like Bill Gates four to six months ago already like explicitly saying he would be pro taxing robots. Yeah. Not specifically AI, but pro taxing robots. This article really goes on, like, should we tax AI if it replaced workers?
Murilo:Yes.
Bart:That's the big
Murilo:if. So what I like about this article is that I do think it's a real problem.
Bart:I do think That there will be displacements, you mean? That is a real problem.
Murilo:I think so I'm not sure if there will be displacement. I think it's like a problem like that as the world evolves, jobs are changing, some jobs don't exist, and maybe some jobs are created. The world changes. Right? And sometimes the thing that with AI, feel like it moves very fast.
Murilo:And so the world doesn't adjust fast enough.
Bart:But if jobs go away, whether it be slow or fast, like, will be displacement.
Murilo:True. True. True. So And the danger
Bart:of AI is that it will go very fast and that it will have a big impact. Exactly. Exactly.
Murilo:And I think I like that there are discussions about this. I like that people worry about this. What I don't like is people say, like, AI, and I what I don't like is that it it kinda improperizes a bit AI Right? Like, if you say, like, would you text, I don't know, computers? You know, like, then it sounds a bit weird.
Murilo:Right? But I I do think, if you look at AI as a as a tool that yeah. Like, any any new technology can can make people productive, can take out jobs, can do all these things, and maybe there should be a text on this. Right? But the only thing they talk in this article I mean, these things explicitly mention is, like, robots and and a so it does feel like a bit, like, anthropomizing it a bit, but I think the discussion should be broader.
Murilo:Like, if there's a new technology that could cause jobs to go away, then should we should we have some some tax mechanism for these things? Right. I don't know what you think about that.
Bart:I think the let's say the pro argument to this is that it could be a good solution, for the for the short term struggle it creates. Because I think if jobs go away and I think it's likely that we will see domains where there is more and more automation and that it will go quicker than we all think. And because it goes so quickly, I think there is something to say to say, maybe we need to cover some of that displacement via taxes. In general, long term, I'm very much against it. K.
Bart:Because it's it's also to me, like, it's super even in short term, it's super impractical. I mean, what is what is this AI that you're gonna text? Like, is it a big provider? Is it someone creating a tool for accounting that uses an alarm somewhere and does some automation? Yeah.
Bart:Like, it's super vague, like who you're gonna text.
Murilo:If you're more productive, like you're not like people are more productive, so not necessarily like people are not losing their jobs necessarily, but kinda is because if five people can do the job of 10 people, you don't need five people now. So how do you tax that? And Yeah. What is like, yeah. I mean, if you say we're gonna text AI, okay, where are you gonna draw the line of what's AI and what should be taxed and what shouldn't be taxed?
Murilo:I also agree that it's a bit of a a band aid solution to, like, you're not solving the the root. Right? Like, it's still there. And, like, there's also the fact, like, okay, if you'd start taxing these things, then maybe companies are less they're not encouraged to use AI. And then it's also maybe not good for innovation.
Murilo:It's not good for, you know,
Bart:if we text the users, you mean? Well, even it doesn't really matter because if if it's the producers, like, will become more expensive.
Murilo:Exactly. Like, one way or another, like, you're gonna discourage a bit. Right? Like so yeah. I mean, we talked about, like, again, referring to the the interview with Charlotte.
Murilo:We did. Right? Like, if you if you slow things down as well, maybe fall a bit behind on the the progress. Right? So how like, it is it is in practice.
Bart:It's difficult. Yeah. Yeah. I'm I'm I'm personally more, like, long term. Like, it's actually mentioned in the article.
Bart:Like, the IMF favors adjusting capital taxes instead. Like, it's such a such a hype domain that there is so much these days being invested in AI that I would not be surprised if we are at this moment, like, every week, we're creating thousands of millionaires. And the reality of that is simply that the the also the article says like 80% of taxes in The US are taxes on labor. Those are not taxes on people that are wealthy and their capital has capital gains.
Murilo:I see.
Bart:And that very quickly becomes very unevenly distributed. Like, even if you look in Belgium, like someone that works every day, the tax on labor, like the tax wedge, like the difference between what you have on on your bank account end of the month and the total cost is like 50% future. And if you have a lot of capital, what you gain through that capital every year, your tax, like, there'll be slightly more next year. And, like, if you mix a bit all types of financial interest, you may be taxed at 20%. Doesn't sound fair.
Bart:Right?
Murilo:No. And
Bart:I think this this you could say maybe thirty, forty years ago like this this we had less wealthy people, but now I think because we save so many huge investments in AI, you're, like, literally, like, creating millionaires at the moment. Like and and this difference will only become bigger. Like, and I think that is a better long term solution to bring that to bring probably tax on capital gains a bit up, tax on labor down, and have a more healthy balance. Yeah. That's true.
Murilo:That's true. That's true.
Bart:And then you also have more money in the system to to find solutions for people that cannot find a job, etcetera.
Murilo:Yeah, that's true. That's true. Yeah, that's true. But there will still be a piece. Right?
Murilo:Like, I think the ideal solution is still to have, I mean, the hope, right. Is that, yeah, some jobs are not going to exist anymore, but maybe there are going to be new jobs or like things are going to be redistribute. Right. And maybe like in twenty years, we'll figure it out.
Bart:Yeah. Long term. I don't think long term, this is indeed something that has happened a lot of times before. Only difference is that it's super fast.
Murilo:Yeah, exactly. And that's also what I, I mean, that's also the first impression I had when I when I was reading this. Right? It's like, this has happened before. This will happen again.
Murilo:I think now there are some differences that it is super fast, but I also what I don't like again, like, I didn't like that. It feels like when you say taxing AI is like you're taxing a little person that you don't see. Right? But it's there. And it's like, you're taxing a tool.
Murilo:You're taxing a technology.
Bart:Yeah. Taxing the use of
Murilo:a technology. True. Right? And I think when if you were to frame it a bit differently, it like, the discussions are also different. Yeah.
Murilo:So that's fair point. Right?
Bart:But yeah. So and I think, honestly, in terms of displacement, I think the robot robots will be a bigger challenge. And I think a lot of people will say it's thirty to fifty years away. I think it's ten years away.
Murilo:Actually, I think so too as well. I think I think you're I think it's it's shorter than most people think.
Bart:I think so. Yeah.
Murilo:Actually. Yeah. I do think it's like, there's a lot of stuff that is possible. I think the main challenge is like, well, one legislation, like, what can these things do? And I think also making it accessible for, like, feasible.
Murilo:Right? I think just technologically, if you go to, like, a very niche lab, you're gonna you probably do some really crazy stuff, but it's not something that everyone can do. So I also agree I also agree with you on that one. What else do we have? This is a interesting one.
Bart:Yes. You're looking at me to read. Pirate preservation group, NasArchive, says it scraped nearly all of Spotify, publishing metadata for 6,000,000 tracks and starting to release 86,000,000 audio files via torrents, about 300 terabyte altogether. Organized by Spotify's popularity metric, the dump claims to cover around 99.6% of listens and will roll out in stages, setting up legal fights over mass scraping. That's big news.
Bart:Right?
Murilo:Yeah. Yeah. It's because actually the the blog so what we're showing on the screen is is a report. Right? Like, on the on this news, basically, but the there was also a blog post attached to it.
Murilo:Right?
Bart:The Well, there's blog post on Anna's archive, but Anna's archive is blocked in a lot of Yeah. Countries.
Murilo:Yeah. Exactly.
Bart:As it is in Belgium.
Murilo:The the blog was also explaining a bit how they did it. No? Yeah. Yeah. So it was also a bit oh, I think from engineering point of view, was interesting.
Murilo:Of course. Like, I understand why they would take it down.
Bart:So what what do mean? Why they would take it down?
Murilo:Like, they don't want to
Bart:You mean why why it's blocked at a Yeah.
Murilo:Why it's Yeah.
Bart:But it was already before.
Murilo:That was already before.
Bart:Amaz Archive is is known for to have, like, almost any book imaginable you can think of. It's
Murilo:It's like Libgen kinda.
Bart:It mirrors a power of Libgen, but it has more more sources than just Libgen.
Murilo:Yeah. But this is I think this is a next level thing.
Bart:Well, it's a different it's a different channel. Right? Like, this is music that they didn't focus on it before.
Murilo:I feel like there's more money in music than books. No? I mean, that's the perception I have. I don't know.
Bart:Yeah. Now I think it it will also put Anna's archive very much, like, on the map again. Like, this is something that needs to be tackled. That is something that needs to be taken down. Like, I think it's for them is a big risk to to For Spotify.
Bart:Yeah. I think it's way more of a thing than than having on than having ebooks there.
Murilo:For sure. I I fully agree. I also think that, like, for ex yeah. Like, I think even at the end of this article. Right?
Murilo:Like, it is obviously illegal. Right? So probably going be a lot
Bart:of lawsuits, lot of illegal.
Murilo:It's, it's not legal. Yeah. Don't do this anyone. But if it's out and it's torrented, it's it's out there. Right?
Murilo:Like people can download and can reshare like the the torrent streams and all these things. Even if they sue and they I don't know if something happens like what happens with Spotify. Right? Because now all the old 99.6% of the things that the average person listens is there. Yeah.
Murilo:True. So it's like, it's, it's, can you put that genie back in the bottle kind of thing? Right. Like, I don't know. So, so, yeah, I'm also surprised, like, did you did you see anywhere, like, any paraphrasing on how they were able to do it as well?
Murilo:Because to me, it's also surprising that they were able to to do this systematically.
Bart:They found a way to get around the DRM, but I don't know the details. No. No. No. It's interesting to see the reactions on, on, for example, Hacker News or Reddit, where they are I I would say they were, for the most part, very positive.
Bart:Ah, yeah. Yeah. Very positive. Because this is a bit like a NS archive, like, they position themselves a bit as a as an activist group.
Murilo:Mhmm.
Bart:Right? We're for the people. We're preserving this for the people.
Murilo:Yeah.
Bart:We're we're, scraping the Internet to, to retain this for history. That's that's the positive thing to to look at it. Right? Yeah. And I think there's something to be said said about it, but I think that is the reason why a
Murilo:lot
Bart:of people do seem very positive about this. But if this would have been OpenAI or Google or, like, like, people would cry outrage. Right?
Murilo:Yeah.
Bart:And Anna's archive is even like they have a they have a page on this, like, they're open to land providers contacting them to basically provide them with data, to sell them data. So it's not it's not I mean, of course, like, this probably costs a lot of infrastructure to run. Yeah. It's not not weird that they're looking for ways to monetize it, but, I mean, end of the day, this is just a huge copyright breach. Right?
Murilo:Yeah. I mean,
Bart:for sure. The question is if it's a problem in a sense that, if I if I look at myself and I'll I'll I'll explain this from, twenty years ago when it was way younger, and all of these things have been expired by now, if I ever did anything illegal. But when I used things like Limewire. Like Limewire, for example. Or what was their OG?
Bart:I forget. I don't remember. The big one before that Napster.
Murilo:Napster. Napster. Yeah. Never used Napster actually.
Bart:Or Torrance or things like that. I think it was it's also a way to know that these artists are out there. Right? If I wouldn't have access to this, like, wouldn't have had the money to buy something in the store at that point.
Murilo:Yeah. I see what you're saying.
Bart:Right? But I was able to appreciate it. And maybe because of that, I went I bought the music later or I played now on Spotify or I went to a concert or I'm not sure if it's actually, like, the grand scheme of things, a problem for the actual artist. Yeah.
Murilo:I think, but I think with the Spotify for artists, I think we've, I've heard a lot that Spotify is not a good deal for artists. It's like, it's a good deal for like marketing quote unquote. Right? Like because people listen to discover them.
Bart:You need to be huge to to really make money there.
Murilo:But I heard that even if you make money, it's like like if you're that big, make way more money by touring.
Bart:Yeah. Of course. Yeah. Yeah.
Murilo:You know? Like or by making TV appearances or whatever right like but like Spotify in itself they pay very little to the content creators you know so I think it's really bad for Spotify but I think maybe the artists they're not.
Bart:And it's also like, it's such a small fraction of your audience that will have the technical know how to torrent this. It's true. Like, I'm not sure how big a problem it is.
Murilo:It's true. Unless, like, someone that is more technical, they're like, I'm launching free Spotify. But, actually, even if you do that, it's like
Bart:Well, what you now will, of course, get is that you have you're gonna get very suddenly, players like Sunu will become very good because they have very rich training data.
Murilo:True.
Bart:True. And they're like, it's the other side. Right? Like, that doesn't feel Yeah. From an ethical point of view.
Bart:Right? Like, they're just
Murilo:Yeah. Yeah. Exactly. But that's the thing. I feel like it's like, it's it's something's gonna change.
Murilo:I don't know exactly what's gonna happen. I think there will be good things. There will be bad things. Yeah. I'm not sure, actually.
Murilo:I'm not sure because we yeah. There's a lot of components here. Right? Because if someone finds another use for this, maybe the use is good or maybe this is not good.
Bart:I think
Murilo:how are you gonna I don't know.
Bart:What I hope that someone builds is an actual recommendation engine based on this. Like, this recommendation of Spotify are shit. Yeah. Like, I've been listening to the same to the same garbage in the last ten years.
Murilo:Spotify reps are like same.
Bart:Horrible.
Murilo:Yeah. But I wonder if it's also a hard problem. I think it can be better than it is now. Well, I do think it can be better, but I'm also wondering if it's I do think it's probably a hard problem.
Bart:Back in the day, had YouTube music before it went to shit, and I think it was way better recommendation.
Murilo:Yeah. Okay. Then maybe they don't have any I cannot excuse Spotify anymore. But, yeah, let's see what happens. For me, I thought it was just I was very surprised to as soon as you shared it, I opened it, and I was reading it, and I was like, what the fuck?
Bart:You opened BitTorrent. You, Torrent.
Murilo:No. I actually I I read the the article that Yeah.
Bart:Of course. That's official. Let's go to the next one. What the hell do we have, Miguel?
Murilo:We have let me share this tab here. But now I introduced a new CHEDGPT images experience powered by GPT image 1.5 promising more precise edits, better instruction following, and up to four times faster generation. It's rolling out to CHETH PT or it has already rolled out with business and enterprise access following and a revamped creation space for workflows. So it's a new, image model now. I looked at well, do you wanna share actually your, your, your test trial?
Bart:Yeah, we'll do that. I'm gonna quickly. So while I'll
Murilo:do this, I'll just go over a bit over the things that they mentioned here. So it's a new image model. Right? You can use with API or on JGPT. Apparently it's better a lot of stuff.
Murilo:Right? Like editing, saying like, hey, put these three images together in one picture. And the example here is like two two people on a and also like iterating a bit, like saying, okay, do this, do that, include this, don't include that. Following instructions. So if you give like a long prompt, it's better at keeping track of all the things that you need to do.
Murilo:And there are also the text, making sure that the text in the images are actually actual text and not just gibberish. Yeah. And some other stuff. There are some regressions that they noted, which I thought it was also interesting. For example, styles, apparently this before used to be better.
Murilo:So the new one, you say draw me like a dark fantasy anime, now it looks a bit like a person's. And before it actually looked more like an anime.
Bart:Yeah. Interesting.
Murilo:Yeah. So there are some some some stuff that used to be better, which I thought it was nice that they were very transparent about it as well.
Bart:In my opinion I mean, my this is the second best image generation model. I think, Nano Banana still the Pro is still better. I've sent you a link over WhatsApp. So we have so maybe if you're a bit into this, like, LM benchmark scene, you know the pelican test. Pelican on a bike test.
Bart:Right? This comes from Pelican on a bike comes from, Simon Willis, I want to say. Is that correct? Maybe.
Murilo:Let me check.
Bart:Simon Willisson. Yeah. And he basically, when a new LLM comes out, he says, draw me a pelican riding a bike. Sometimes it's good, sometimes it's horrible.
Murilo:To this guy. Yeah. Simon Woolizons. Yeah.
Bart:Yeah. And what we've been doing, really Milo really asked me to do this, over the the course of the last last three, four versions of image generation models, so I what we do now, our protocol, is that I paste in Merilo's image. Maybe you can share it on the screen. Do you have it?
Murilo:I'm just building the courage to. No.
Bart:No. That's if you go up if you go up a little
Murilo:bit. I cannot.
Bart:Oh, you cannot. But this is basically like this image of Merilo, if you download it, There's only one image on the whole worldwide web of Marillo that you can find. This is LinkedIn image. Right. That's the one I use.
Bart:And then you give this instruction, put this guy on a white horse, shirtless. The guy has long white hairs and is eating a banana. And this is what comes out. And this is honestly, like, it's on par with Nano Banana Pro. Right?
Bart:This was my very first test with the new image model of Shiji PD. It is pretty good. It is pretty good.
Murilo:It is pretty good. I still think like, again, I'm a little bit picky about that. So we talked on the banana banana and we explain how every time we try to generate something with me and Bart, Bart's like, And with me, it's always like, not me.
Bart:Well, it has very long been the case. Like, if we try to do something with you, you either turn, like you become Chinese or you become Pakistani or
Murilo:some like nothing. It's not me.
Bart:But but it's also like it to me, like, it's a bit like like an NP hard problem. If you would say to someone out in the street, like, draw this representation of the guy, but keep in mind that this guy is half Japanese, half Brazilian. And make it look like a half Japanese and half Brazilian. No one knows what to draw. Right?
Murilo:That's true. No one knows. Maybe it'll be like Pakistani or something.
Bart:But this is the first time like, actually, Banana Banana was the first time
Murilo:Banana Banana was the first time.
Bart:Banana Banana Pro. The regular one, not even that.
Murilo:Yeah. Exactly. True.
Bart:Nano Banana Pro is the very first one.
Murilo:But I think it's like, when I look at this, I I can see myself kinda, but like It's recognizable. It is recognizable, but I don't it's not like if you pay attention
Bart:You've drifted off to Mexican a little bit.
Murilo:Yeah. A bit. Right? Like, it's I don't know. Like, I don't think I look bad.
Murilo:Don't know. But there's some about it that, like, it's kinda like me, but kinda, you know, it's a bit different. I feel like the other one, the nano banana pro one, really like. That could be me. You know?
Bart:It Yeah. That you could send to your mother and say
Murilo:And it'd be like, hey. Like, woah.
Bart:I was riding a horse out
Murilo:of the
Bart:fields in Belgium.
Murilo:Exactly. And I was after I got my nationality
Bart:Yeah.
Murilo:I just turned blonde for some reason. Know? It was like, this one, I think she would be like, you know, I think this could be like a a brother of mine. You know?
Bart:Okay. So our how do we call this test? We can't call it a pelican test. How about you come up with
Murilo:the name for this one?
Bart:I'll think about it. But on our test, Nano Banana Pro is still, the best one.
Murilo:Yeah. But I would say like, if Nano Banana is a 10, this is like a solid nine. Yeah. It's close. It's very close.
Bart:And I think it's also, it's a hard enough test because it's like, start from a picture where you're just clothed.
Murilo:Yeah.
Bart:And then you say put him on a white horse, eating a banana, long white hair, shirtless. Like it's a lot of lot of things.
Murilo:There is a lot of things. There's a lot of things. I mean also Che Gpt made it a bit hard on itself. Like changed a bit the angle and, like, also added some some wave, like, you know, wind on the hair and stuff.
Bart:I wouldn't be impressed if this guy is just I'm, like, strolling just strolling down the street. I'm sorry.
Murilo:Yeah. But I I like that all all of them, they they kinda give me, like, a good physique. You know? This
Bart:is not your physique?
Murilo:No comment.
Bart:It's going in the direction. Right? I heard some rumors at some point point about about apps. Yeah. Else do we have?
Bart:Charlie March of Astrofame announced the beta of Ty, I guess it's pronounced, t y, an extremely fast Python type checker and language server written in Rust, aiming to speed up feedback for large code bases. The release invites early testing and comparisons, signaling another push for high performance tooling from the creator of rough and UV.
Murilo:Yes. So I think this is the first. Astral tool. So don't get me wrong. I actually quite like the the Astral tools.
Murilo:I like rough. I like UV.
Bart:You like rough. Okay. Sorry. Okay. Let's continue.
Murilo:Let's keep the
Bart:show. Yeah. Sorry. So let's keep professional.
Murilo:I like the tool rough. Yeah. I like UV. I haven't tried Ty yet, but Ty is the first tool that I'm like, that would be nice if we had a tool written in Rust about this. Like, he's the first tool that I'm like, performance is an issue.
Bart:Maybe explain a bit for the audience about about what is the type checker.
Murilo:Yeah. So a type check so basically in Python, things are not typed. Right? Like, everything's dynamic. So when you try like, if you have x is equal to blah, blah can be a number, it can be a string, can be whatever.
Murilo:And it's not gonna complain to you when you when you try to run it. Right? What it will complain is if it's actually a number and you try to do things that you want like, put I put this number in uppercase, then it's gonna be like, you cannot do this. But it's only a runtime. Python, they released also typing.
Murilo:So basically, it's something that is more for people to see. It it doesn't influence necessarily the runtime just to say, hey. I'm I'm creating this this variable x, and this is gonna be an integer. Right? And he's, oh, it's an integer.
Murilo:Okay. Cool. So I'm gonna know. Maybe informative. It's informative.
Murilo:Right? But that's why they even call it hints because it's just like a hint. Even if I say, oh, this number is an integer, but actually I write hello world. That's still okay. Like the program still runs.
Bart:You could theoretically pass might at run time might give an error
Murilo:might give an error at
Bart:the time. Program will start.
Murilo:Exactly. Exactly. Basically have static type checkers. Static, because it's not when the program is running, it's like, it just looks at the files and says, hey. This is a number, and you're trying to do it later on.
Murilo:You're trying to do something that you do in a string. There's an error. Or maybe you're trying to add a number and a string. That's not okay. And these things can get actually very complicated because a lot of times you see variables that are like, ah, this can be a number, a string, or anything that has this method.
Murilo:Right. And basically there are type checkers that kind of go through your whole program and they see how the things interact with each other. And they see if there are any violations, anything that anything that could go wrong. Right. And then it gets a bit tricky.
Murilo:So for example, maybe there are functions that they can return none, which is a type in itself, but they never do really because the way that your program is called, the way it's set up, that would never happen. But because the function is set up like this, it can actually break stuff down the line. So it can get very tricky, and the the the very popular one, I think that maybe the first one is Mypy.
Bart:If you're a Python developer, you've probably heard of MyPy.
Murilo:Exactly. I think I think so too. Then since then, there was, like, Pyrite, which I think is from Facebook.
Bart:No. The one above is from Facebook. Firefly. Oh, I want to say Pyrite is from Microsoft. I'm
Murilo:actually think you're right. I think you're right. I think Pyrite is from Microsoft. And then Pyrite came out actually, like, close released, like, their first announcement. I think I heard it close to Ty as well.
Murilo:And Ty is like the new one from from Astro and it's written in Rust. So it's actually very fast. And Mypy is the first of the Astro tooling that I'm actually, I see the point of writing something in Rust because even when rough came out
Bart:Because mypy is slow on large code base.
Murilo:It is. And there's they try to do a of caching, try to do a lot of stuff like this, but the reality is it is it is slow. When rough came out, so rough is like a linter. Right? I was like, okay, it's cool.
Murilo:But I never really had that as an issue. Right? Even though now I use it and I do now that it's faster, I do feel like I would miss it if I went back. Same thing for UV, but Thai, I think it's one that a lot of people already felt the pain before. So I'm happy that they're sharing this.
Murilo:This is still beta, meaning that you shouldn't look at this as like something production. You should expect issues. Right? But it should have a good coverage with with mypy. Maybe I'll share this tab instead.
Murilo:And one thing that I also saw that I thought it was interesting is that they even released they even have, like in the type system, they actually add a few things that is not in Python. So it's actually a superset of Python type hints. Where is it? There was one thing about, like, for example, intersections. If you want something to be A and B, that's not something you can do with Python.
Murilo:Like just the built in things. Yeah. Intersection types. And actually, where is it? Tye?
Murilo:Yeah. Here. So they actually added, like, an tie extensions. You can import intersection, and you can create a new type that is an intersection between two things. So basically, my thing is one and the other.
Murilo:So tie is not just a replacement of mypy. It doesn't just work with the the standard type ins, but they're also adding new things on top of that, including like a language server. Right? So one of the nice things about type ins is that if you're in the in the IDE. Right?
Murilo:And you hover over variable, the IDE can tell you more information about this and that's done by a language server. So they also release a language server with this. And again, language service, it's very speed is also a big deal. Right. So I'm also quite, I'm quite excited about this.
Murilo:Let's see.
Bart:Yeah, it's cool. I'm sometimes there's a very large potential user base for this. Well, Python developers. Right? But at the same time, it's still a very niche tool.
Bart:And in that sense, I'm I'm sometimes I wonder, like, would these teams would it not be better for them to cooperate?
Murilo:Corporate how?
Bart:Like, you have Microsoft, you have Meta doing this, you have Estral doing this. Like, should you not put a few more acts in a few less baskets?
Murilo:Yeah. I mean, I see what you're saying. But maybe at the same time, it's like, I wonder if they they have this for example, like written in Rust or written in another language. That's a big decision. Right?
Murilo:And maybe spreading a bit, you also reduce the risk. Right?
Bart:Yeah. Maybe. Maybe.
Murilo:I'm not sure. One thing that you mentioned when I shared this was monetization from Astro. Right?
Bart:Yeah. I'm very curious what their path to monetization is.
Murilo:So Astro maybe for yeah. Astro is the company behind all these three tools that we mentioned.
Bart:And they raised from memory, like, I I want to say between eight and twelve million two ish years ago.
Murilo:Yeah. So
Bart:I mean, at some point that needs to turn into some monetization.
Murilo:And everything they're building now is open source.
Bart:Everything they're building now is open source as far as I know.
Murilo:As far as I know, like even like you can use it, whatever. You can do whatever you want with it, basically.
Bart:As far as I know, they they also don't do any consulting. Maybe I'm wrong there. Yeah. But at some point they do need to monetize. Like, will it be will it be consulting?
Bart:Will they build a service around tooling, which I think is super hard to do? Because then you're basically gonna say we're gonna set with our tool set, like, we're gonna make CICD for Python, like, super easy, super good developer experience, but, like, which enterprise? Because you need an enterprise to actually buy such licenses, like, which enterprise is gonna buy something specifically for Python. There's already a very small group. Maybe they will sell off some of their assets to someone else that can monetize.
Bart:It has a big, bigger ecosystem. Yeah. And I I don't know. Like, I'm very curious to say to see what they will do five years from now.
Murilo:Well, I do think it's good for the community in any case. Right? Because these things are open source during GitHub. So I feel like for even if they don't have a happy path, let's say Yeah. There's a lot that is done already.
Bart:Oh, yeah. Definitely.
Murilo:Which is good. But I agree with you. I think maybe But
Bart:it would be a pity if it's if the company at some point fails because they don't have a part two monetization.
Murilo:Yeah. Yeah. Yeah.
Bart:I see. Because it like you're saying, we're doing a lot of good things for the community.
Murilo:Exactly. And the stuff that they're building, they it looks good. Like UV, you feel like it's a bit the standard rough.
Bart:Very quickly became
Murilo:Exactly.
Bart:A lot of the tools very quickly became the standard.
Murilo:Exactly. So it's like it's it's a it's a good they seem to be well designed, well thought out, you know, so indeed it would be a pity. Maybe one other thing that I just mentioned here, like, just saw that they also have a py Pyx, so pyx, Python native package registry. Oh, that might be interesting. Yeah.
Bart:I didn't know. This is also very like, August 13. I also missed this.
Murilo:I missed this as well.
Bart:That could be a good path to monetization. So if they have their own native package registry, that's something that they can offer to to a large enterprise, like an internal registry that is safe and tested. And So maybe It's a bit maybe the Anaconda's route then. Yeah. Yeah.
Bart:Indeed. With a bit bit of a different approach because but
Murilo:Yeah. So, you know, maybe here on the announcement, PyX is the Python native package registered in the first piece of the Astro platform, our next generation infrastructure for Python ecosystem. So I think this this probably can shed some light. So, yeah, I think it could be a more secure GPU aware, private packages, public sources. So,
Bart:you know, could
Murilo:be could be something interesting in
Bart:the end. Curious to see what they will do because it's not easy to find. I think product market fits with like a infrastructure for Python.
Murilo:Yeah. Yeah. Yeah. Platform. Yeah.
Murilo:It's true. Especially when you have like tools and like, it's not a service. Right? It's just like something just
Bart:Yeah. User. But yeah.
Murilo:But I think also becoming the standard for Python ecosystem, I think if you really
Bart:Like, said like this there's I can't mention another company that has in recent years has been done so many things that have all become the center in the community. Like they have a huge, like a very strong track record. Right? That's true. That's true.
Murilo:And I think also if you're using your view, using all these things and you have the option to go for these different package registries. Right.
Bart:I think it's So if anyone can get surprises, would I go to market as them, I guess.
Murilo:Yeah. So, yeah. Cool. I mean, we'll keep an eye out. Right.
Murilo:Let's see. Let's see what happens. And next, what do we have? It's my turn. No.
Bart:Yes.
Murilo:So next, we have alpr. Watch maps, city and county meetings where automated license plates, readers, facial recognition, or flock safety appear on agendas so residents can see surveillance decisions in motion. It scans public documents for keywords and pins them on a live map with options to report cameras or get local alerts. So what is this about Bart?
Bart:Well, this is a link I I came across on Hacker News. ALPR stands for
Murilo:I think it's automatic license plate recognition.
Bart:Yeah. Well, apparently, there's a whole vigilante group that is very against the automated license plate recognition. I didn't know that this was something that we were angry about. Think this is something that just happened in Belgium. Right?
Bart:Like, one really I think so. No one really
Murilo:I think so.
Bart:No one really questioned this for some reason. But of course, like they they they they can know wherever you are when you ride some when you drive somewhere.
Murilo:Yeah. I think also not just that. Like, if they take pictures and you have someone in the car like, actually, like, I think in Germany, they just send the they send the picture as well to your like, with the like, if you go over the speed limit, they take a picture. They send you the picture as well.
Bart:It's always a bad picture of me. Yeah. But it's like I'm also Making my nose. Or
Murilo:Sleeping. Yeah. But, like, it can also be a bit like, I don't know. Right? Like, what if you're I don't know.
Bart:Who did you have sitting next to you?
Murilo:I have Maria, like my wife, but like, they had a they had a
Bart:That was a very quick answer.
Murilo:Sure. But like, because actually for the picture, had like, they cropped the the box. Right? So I thought they always did this, but I was talking to a colleague and she was saying like no it's a big issue because she was driving she was supposed to be doing this because the parents didn't know when the car was registered at the parents house and they said it was like it was this whole thing Right? Which again, like, maybe it's not the biggest deal.
Murilo:I mean, there are also use cases here and there or maybe some circumstances. Right? Like, it's a bit sensitive.
Bart:And the website is also broader. I think it started from LPRs. Now they do they always focus on facial recognition. Block cameras. I don't know what they are.
Bart:Me neither. Basically, any surveillance tech that's potentially, like, decreases privacy. What I found interesting is actually the way that they do it. Like, it's a bit like their their their counter attack is also using data to make things visible. Sure.
Bart:And a very smart way. Like, they use meetings from local councils and do a lot of automated data exploration to see, like, if any of these measures are being discussed and what the discussion is to basically be aware and so that you can also if if you are aware, you can you can, if you want, influence this this decision making process. Yeah. I thought it was a very interesting way to do this and like translating that to Belgium. I don't even know if if council meetings are public here.
Bart:The reports of that, like, don't know. I've never seen anything being done with it.
Murilo:Yeah. I'm not sure.
Bart:I I thought this, like, in general, I'm when it comes to open data, like government publishers are always a bit like underwhelmed of what you can actually do with it. But like, this is to me is a very strong use case. Like you can actually learn something. You're like, you can you can learn here how policy around these topics is being formed and what stages they are like. Yeah, This is a very smart way to, to tackle this.
Murilo:I think also when they did this, so I mean, I'm showing here a map, right, when they have a whole bunch of like gray exclamation marks, which was where this was discussed in the past, which maybe by in itself, it's already interesting to see where this discussed more than other places. But you see, like, what's current, what's in the past. But I guess, like, they they wanted, like what I understood is, a bit to promote a bit, like, people protesting. Like, if you don't believe this, like, go over your neighborhood and, like, try to to pressure the politicians to to not pass these things. I think I saw, and I'm not sure if I'm making this up or if you show me or something.
Murilo:But there were some of the thinking the European commission, some sessions that are live streamed. Yeah. Right? Yeah. And there was like a system that was to check if the politicians were on their phone.
Bart:Oh, yeah. And I didn't show you, but
Murilo:And so like and and then actually, like, if someone's on their phone, it was automated to to to tag them on Twitter, say like, and with the screenshot. Oh, nice. It's be like, hey, stop. Like, don't be on your phone anymore. Like, I I I need to find it.
Murilo:Maybe I'll share it like next time.
Bart:But like, I thought
Murilo:that was fun. That was also neither, like another another fun use of this of this data. So, yeah, really cool.
Bart:This would be better for a picking nose. Yeah. Yeah. Yeah. Think.
Bart:Clean
Murilo:your nose, clean your hand, you know, it's like, yeah, that would be So, yeah. Cool. Cool. Cool. And a lot of, a lot of different stats here.
Murilo:Right. What else? What's next Bart?
Bart:We have two more. OpenAI. Yet again. OpenAI opened app submissions for ChatGPT and launched an in product app directory, putting third party tools a click away in conversations. Submissions are reviewed for safety and quality.
Bart:Link outs are allowed initially and the first approved apps will roll out in the new year as monetization options evolve.
Murilo:So what are these apps on ChatGPT?
Bart:I haven't used them yet. And there are for example, Lovable already published their app, and you can basically use them in the sense that you can add them. Like, you can use an at sign and then say, at Lovable. Can you build this prototype for me?
Murilo:In your ChatGPT conversations. Okay.
Bart:Yeah. Interface. Or you can say I'm just thinking out loud here. Or you can say, at Amazon, I would like to have this this delivered by Thursday.
Murilo:I see.
Bart:And to me, this is huge news. It is? This can change the Internet. Like, to me, this is their big bet, their big platform bet. They have 800,000,000 users that have ChatGPT installed.
Bart:True. If they can find the right offering, and I think a lot of companies are willing to develop apps for ChatGPT because it's a potentially use user. I don't think it's too far fetched that this will be like ChatGPT will become like a huge channel all on its own and people will not have to leave ChatGPT.
Murilo:True. That's actually very true.
Bart:And I was looking at the stats like apparently in the world because we we often think or I often think about like how are you going to do to the Internet? And then of course you discuss like you have browsers, but browsers are more or less always the same. Right? Like there's not really a big difference. But then you took a talk about like, device.
Bart:And then it's like, it's either your computer or your mobile or tablet. Right? To me, this could be like something new. Like, you only use ChatGPT.
Murilo:Yeah. Like, that's your way to interact.
Bart:That's your way to interact. And I think, like, for us, it will probably not be the case because for us or not, the audience that will just do that. Right? Like, we will we we use our tablet and we use we use our computer, and we use our mobile, and we use all these different channels, and we use Chattypedia just another app. Right?
Bart:But, apparently, in the world, there are I looked up these stats. There are 40% of people that only access the Internet via their tablet or mobile. 100%. 40%.
Murilo:Oof. That's a lot.
Bart:That's a lot. Yeah.
Murilo:Oh, wow.
Bart:And, like, I think that ChatGPT, even even even though it's not a device, like, it can be, like, a concrete new channel for potentially a big audience. It's true. Because if if the main thing that you do is on the Internet, like and I'm ignoring social apps here. Right? I'm ignoring social.
Bart:And the Internet is the world where where you like, maybe you need read news. Maybe you, do do some shopping. Maybe you check your emails, like theoretically, you could all do that. Right. True and true.
Murilo:Very true. I'm always wondering if social media to what extent could you also?
Bart:Yeah, potentially.
Murilo:Yeah, it's true. It's true. There's all you can do there. And I feel like, yeah. And then how the monetization would work for Chagpty in this scenario?
Murilo:Is this is this transparent? Because
Bart:Well, to me, that's that's relatively easy in a sense, like, but but it all stands of force. Like, can they get a a large enough user base on this that is dedicated enough? Because if they can, it's a simple monetization as the app store does for Apple or the place or does for Google, where Apple goes as far as saying 30% of every transaction is for us, which is huge. Right? Yeah.
Bart:It's huge and like everybody thinks it's way too much, but like no one has a choice because the users are there.
Murilo:Yeah. Yeah. I searched it, but then like, do you think there's also the risk of
Bart:this is a good I'm not saying it's a good thing.
Murilo:I'm not saying it's
Bart:good thing, by the way.
Murilo:Yeah. Yeah. Yeah. But, like, let's imagine that this strategy is successful for for OpenAI, and Gemini starts doing this. You know, like, I'm I'm wondering if
Bart:But no one knows Gemini. You and I know Gemini.
Murilo:Yeah. That's true.
Bart:But the the the the random person you see in the seat doesn't know Gemini. Maybe they heard it now because none of banana was on the news.
Murilo:Yeah. I know what you're saying. It's true. It's true. I'm wondering, like, still, like, on the the I mean, I see your point, and I I do think there's a lot of potential, and I think this could be a very big way for OpenAI to to bring money in basically.
Murilo:But I'm also wondering, like, what's the growth of that? Like, how do you keep how how do you stay ahead in that in the this is how people use the internet because things will evolve. Right? Like things will change. Right?
Bart:Yeah. Well, it's a big bet, of course. But you could say this like, Amazon created e commerce. That's true. Maybe maybe OpenAI will will will create the AI powered Internet.
Murilo:Yeah. That's And, again, I'm not saying I'm
Bart:a good I think it's a good thing. Yeah. But I think, like, if one of these players could do it, I think they could do it.
Murilo:Well, if yeah. I think OpenAI is the the the best position when we do this today. Right?
Bart:And it's clear, like, what what the steps that we're taking is clearly the direction that's thinking. Like, we're all already seen, like, closer integration of of, I think it was Shopify. Right?
Murilo:Like Shopify. Commerce protocol. Yeah. It's true. Yeah.
Murilo:It's true. It's true. It's true. I was listening on podcast, actually, someone that was using Sora, the social media thing. I think we talked about this, how they wanted to create Sora, like a social media platform where everything is just generated by Sora.
Murilo:And actually the the the podcaster guy, like, I think he was being interviewed or something, but, like, he was actually very positive about it. Oh, yeah. He he said it was really fun because also he like, well, the thing that bothers him is, like, when people are trying to sell AI slop as original content, but, there everything is is AI generative.
Bart:Yeah. Yeah. You don't have this dichotomy of, like, like, is this authentic or is this being faked or is this
Murilo:Yeah. Which which takes a bit away to, like, person's trying to trick me vibe.
Bart:Yeah. Yeah. Takes it Vibes like the Like,
Murilo:it's all transparent. That gave a nice vibe.
Bart:That's a fair point, actually.
Murilo:You know, it's like it's about creative, about being goofy, about, like, you put yourself doing funny things or put yourself in a movie or something, you know? So it actually, like, that person was really, really positive about this. Still not a 100% convinced, but maybe actually, like, and I think we have one more, maybe because it's a bit linked with I don't know if you put it on the tidbits with the flamingo picture.
Bart:Oh, no, I didn't.
Murilo:You didn't, but maybe we can, like, it may be in line with that. Right? Like, so basically someone, they entered a photograph AI generated competition
Bart:for images. He actually A competition for AI generated images?
Murilo:Yes. But his entry was actually a real picture of a Flamingo.
Bart:And it was a picture of a Flamingo without basically a neck and a hat. Yeah. But it was like the it had its its its neck tucked in under its wing on the other side of the body, basically.
Murilo:Yeah. Exactly.
Bart:But it looked like a flamingo without a hat. It looked very AI generated.
Murilo:Yeah. And it's
Bart:actually a little She won. I think it was a woman.
Murilo:Oh, okay. Let's see if I can find it.
Bart:And, well, the at the end, they were disqualified, which also, like, was okay because, like, they broke the rules. It was a bit, it's interesting to see these, these things.
Murilo:Yeah. I found it here. And it was like also
Bart:It was on purpose. It was a bit to trigger like a thought process. Right? Exactly. What is creativity?
Murilo:It's a bit like almost like a not protest. Right? But like a bit like everyone's all hyped up about AI, but
Bart:A a tongue in cheek protest.
Murilo:Yeah. Exactly. I mean, lighthearted. Right? But it's more like and so I'm just showing the article here for people.
Murilo:Right? And it kinda shows a bit the story, okay. I was doing this, and I saw this, then I took the picture and yada yada. Kinda like saying, okay. Even though we're in this AI world and AI is all the hype, first, real pictures are better, and second, people still crave a bit that, you know, like, there is still something about the real pictures that AI cannot recreate.
Murilo:So I thought it was quite I thought it was quite refreshing as well. And, yeah, thinking of social media and soar and all these things, I think it can be fun, but I also think that it will never take away from the real content, right? Not the the AI slob. Alright. And last but not least, have Toad.
Murilo:Is it my turn?
Bart:I think it is. Yeah. Think it's my turn. Let's say it is.
Murilo:So Wilma Googan unveiled Toad, a terminal app that unifies AI coding agents under a single UI using the agent client protocol, Aiming to make agentic workflows feel native to the shell. At launch, it supports 12 agent CLIs and adds niceties like markdown streaming and fuzzy file inserts. So Wilma Gugun, he's the guy behind Rich or the guy that created Rich?
Bart:Rich and Textual. Right?
Murilo:And Textual later. And he actually had a a
Bart:start Which is very famous for Python based TUI's. Exactly. TUI's like Textual user interface.
Murilo:Textual user interface. So it's like basically a UI in your in your terminal.
Bart:Yeah. Yeah.
Murilo:Right? Actually, two is terminal user interface.
Bart:Oh, thank you.
Murilo:Yeah. Yeah. Yeah. So basically, it just kinda introduces Thoth, which actually is a wordplay on textual code.
Bart:Oh, I didn't know that.
Murilo:Yeah. That's what he put there. The way I saw this is kind of like a idea, but for your terminal.
Bart:Well, the way I saw it. Oh,
Murilo:did you say it?
Bart:Is that this is an alternative to cloth coat or a codex or a Gemini CLI or Vibe from Mistral, but then provide agnostic.
Murilo:Yeah. But that's also why I thought like of IDE a bit, you know, like for the genetic coding kind of thing, you know, so it's like you have the same.
Bart:It's a richer interface indeed than
Murilo:the old time. But it's like if you like you can the same, how can I, for example, you can use Visual Studio with Claude, You can use Zed with GitHub with Copilot GitHub Copilot? K. Like, there are different IDEs that you can still have the same models behind. Right?
Murilo:So in the end of the day, it's not that the the model is better. It's more like the experience around it. Right? Like, is it intuitive that this pops up here when I want to add this to context, like cursor, Versus Code, like all these things, like, they they have the same capabilities AI wise, it's just a different interface. Yeah.
Murilo:Right? Which which can make a big difference as well. A different Tui. Exactly. And I feel like this is kinda how I felt this, but it's more narrow.
Murilo:It's for agentic coding, so it's really just for Cloud Code and all these things. But you can swap as well. So you can do Gemini CLI codex, cloud code.
Bart:Exactly. You basically have the single UI across multiple providers.
Murilo:Exactly. And, so there, I mean, there are a few things that he also added, right? Like, he wanted to make this feel a bit more like the browser. So if you put, like, the exclamation mark, that's just a shell command, basically. If you have these dynamic applications like htop, right, which kinda shows the the usage of your PC.
Murilo:You can actually do this in Toad and it won't break anything. Like you will parse everything nicely. The images will be there. If you want to, to add file, also auto completion as well. So if you want to run a command, you can just like tab and it will complete for you.
Murilo:So there's some, there's some nice things. Maybe the last one here, he kind of borrowed this from a notebooks as well that you can go up on like different, it's not a cell. Right. But like different conversation messages and you can actually cop stuff to your clipboard. Right.
Murilo:So there's some nice things. For example, markdown formatting as well. They're like, it also formats the markdown as you're typing. So it adds nice colors and all these different things. Some interesting stuff.
Murilo:I was also wondering because for you, you mentioned one time that you think that agentic coding, you should not use the ID. You should just stick to the terminal.
Bart:Did you
Murilo:say that? No?
Bart:I said that.
Murilo:You said that. How do you feel about Tote, or do you feel like this is not really that useful? Maybe one last thing that I think it could be interesting. Yeah. Like, if you also wanna add files to it, can just use the at, and you can actually parse through the whole, like, the the tree.
Murilo:Right? To find the files, it's it's easier to inject in the context. And maybe one thing as well that he he added, you can actually use your cursor. So if you wanna click on stuff, like click on directories and stuff, you can put that on the terminal, which
Bart:Yeah. And difficult to answer this. I think it's good that it exists. I think it's always cool to see what, what we'll build. I think it's you can learn a lot by using this tool by exploring different providers to see how they function.
Bart:I'm not a fan of making the two way too rich, maybe to give too much options. I think it gets cluttered very quickly. Very much a fan of, at this moment in time, at least how how something like clot code or codex works, which is quite minimal. Right?
Murilo:And they're kind of a bit the same. Like, it's not like if you go from clot codes to codex that it's very different. The experience
Bart:is like Exactly. Exactly. More more of a gripe with Gemini. This may be also something that that that McGugan here gets annoyed by. It's like Gemini, if you copy paste on it and stuff like they they've ASCII signs in like on the sides and like it's it gets gets messy quickly.
Bart:But it's an easy fix I get. The thing is, you are willing to pay for the provider at this point, you're gonna go for the best one and you're at this point in time, you're probably gonna go for clot.
Murilo:Gemini CLI is if is there there's a is there a nice free tier? There is no for Google users.
Bart:I don't know if there is a free tier. Well, what there was not one when it was just released, Gemini three. Okay. I don't know. Maybe there's a cheaper model that they can use for free.
Bart:Not but not the safe yard one. So the CLIFE or Cloth codes, like, go for there. I would go there because it also, like, if there's any proprietary integrations that only Claude provides, it's very hard to support it through an open ecosystem. So, yeah, I'm I'm I'm a bit I think it's good for the community that these things exist. I don't see myself switching very quickly, but I'm also not like the guy's of course, passionate about TUIs.
Bart:He's also, he talks about what he calls terminal yank, like that you have these flickering and all these other. They flicker and the redraw approaches.
Murilo:Yeah. He's very passionate about this.
Bart:I've never seen it. Right? Like, they're like, it's it's not a thing.
Murilo:Like, it's also like the different colors, all the range of colors and all these different things.
Bart:And it's very good that we have these people because they do bring us forward. I honestly believe that, but I don't think that I personally will very quickly switch to this tool.
Murilo:But also like, because you and that's also what I was curious because you like the like when you said agent decoding should be on the IDE and not, no, not in the IDE, terminal and the CLI and not on the IDE, but it is because there's less stuff. There's less clutter on the way.
Bart:Exactly. Yeah. Yeah. Exactly. Yeah.
Murilo:So it's like, if you add too much stuff on the terminal, it kind of defeats the point of going to
Bart:the terminal for you. For me. And another issue kinda linked with that. I have an issue that I have with TUIs and they would try to do too many things. You need to start to learn so many shortcuts and it's very specific often to that.
Bart:So it becomes like, if you want to be a power user in the tool, you can be very good, but you need to be dedicated enough to become a power user in the tool.
Murilo:Yeah. Best case you you you turn into Vim.
Bart:Yeah. Something like that. Yeah.
Murilo:No. But I I I see what you're saying as well. I think
Bart:that And Vim is super strong if you are power user. Yeah. And otherwise, it's like as simple as like as like a nano as an adult. Right?
Murilo:Yeah. It's true. I mean, maybe you take a bit longer, but also, like, how often are your nano that you're reading to be dead? You know? It's like, yeah, like, know Vim can you you can go quite, quite, quite far with with Vim.
Murilo:So, yeah, that is it for today. We do have some some tidbits. Do you wanna go over them, Barto? Should I?
Bart:Yeah. Go. No. Let's let's keep short because I limit Yeah. Please for time.
Bart:Right?
Murilo:We'll keep it short. So maybe skills. So that was, that was called skills. Right? Like, was released by Entropic and now Entropic released as an official standard, quote unquote.
Murilo:So it has its own its own website. Right? There it's maintained by Entropic, but it's meant to be for the community similar idea as MCP. And, actually, I wonder as well if, like, skills is becoming the new MCP. Like, feel like a lot of people actually prefer skills over MCP.
Murilo:Like, it's a better design. So I'm curious how that's gonna be that's gonna evolve, especially because MCP was donated to the Linux Foundation and all these things.
Bart:Yeah. I think what we will see, so maybe for, for people that are not familiar, so skills is like really like a written description of how to perform a task.
Murilo:Yeah. And you can have some code as well attached to it, you know?
Bart:Yeah. Can have some code as well. And it can also say like, if you need to generate a PDF, use this code that is attached and use it in this way and call them that way with these parameters and execute it there. Right. With MCP is basically a wrapper around something.
Bart:And I think what we will see towards features at MCP will I think where MCP's are very valuable is like as a wrapper around an API, random API, like programmatic API of a SaaS or whatever. Right?
Murilo:Yeah. Yeah.
Bart:Yeah. And for the more local or simpler or or also more complex things, like, that are that are basically not an API, I think skills will be used.
Murilo:Yeah. I think it's like if if the stuff you're trying to do sits on the other side, like we saw MCP service for Snowflake or for, like, the London Stock Exchange. Right? You cannot do skills with that. Yeah.
Murilo:Right? Like, you need the data's done that side. Like, all the infrastructure, the services there.
Bart:Good way maybe to look at it. Yeah.
Murilo:Right? And I think
Bart:theoretically, could, because you could describe it because of it, but it becomes very complex to set up and
Murilo:Yeah. I think so. I think like if something needs to run on that side, then the server, like MCP server makes sense. And if it's just like a function, if it's just something like this, then I think it makes sense to have skills.
Bart:So I
Murilo:think they can coexist there, but I also think it's nice that this again, and again, open AI also adopted this officially. So another big step there for this.
Bart:Exactly. It was rumored with 5.2 that they used it and now they officially announced it. Yeah.
Murilo:Yes. So so very cool. Next we have Claude for Chrome. So this was released.
Bart:This was released yesterday, I want to say, or the day before? Not the day before.
Murilo:I think it was the day before. Yeah. Let me check.
Bart:And, it's basically a browser extension, a Chrome browser extension where you can let, Clot control your browser. And I was hopeful, but it's as shitty as all the other AIs in the browser.
Murilo:I was hopeful also because maybe it, like, uses the DOM and maybe the DOM is code and the cloud models are better with code. That's like except the whole Yeah. That that delusion I told myself.
Bart:So I actually had a use case, so I thought, oh, cloud release something. It must be good. Right? I immediately installed it, and I, I had, I can discuss whether this is private privacy sensitive or not, I was risk with what I did. But I asked it to check my Gmail, because I had a meeting on Monday morning, today, but actually this morning, in Leuven, and I wanted to to verify what time it was.
Bart:That is in my head. I I thought you should check it in my agenda, but actually, I just installed blog post. Let's try. Yeah. I'm not not getting like, it tried for thirty minutes.
Bart:It very quickly opened Gmail, but then, like, it couldn't find the mail. And then it found the mail, but it couldn't find like, it's a threat. Like, it couldn't find the message in the threat. And yeah. Fuck.
Bart:Right? What? Yeah. Like, yeah. No.
Bart:Right? Yeah. No. I really was expecting more. Yeah.
Bart:But again, like, it's, of course, it's a sample size of one, but it's not like, it's not the sample size one. It's not encouraging.
Murilo:Yeah. I mean, for sure. I think we need a pelican on a bike for one as well to try for them, you know, like, is a, but what is a good task? Know? I
Bart:need to think about that. Yeah. I want you to think about this, but
Murilo:maybe we can do a quick, quick rundown, you know, just like
Bart:open the three browsers. Privacy sensitive to every time, like, let them go through my journey.
Murilo:This shoe, this shoe, shoe. Who think of something? Who think of something? What else do we have? We have lovable raised a lot of money again.
Bart:Lovable raised
Murilo:$330,000,000 at 6,600,000,000.0 valuation.
Bart:Yeah. Crazy amount. Yeah. Impressive. I very much wonder where Lovable is five to ten years from now.
Bart:Let's say five, ten years from now. Very what they're what they're very good at, you can't deny them that, like, they they they are extremely good at go to market. They were viral prototyping tool. If you very quickly want to build an app, if you're not a developer, you know of Lovable. For sure.
Bart:They're extremely good at marketing.
Murilo:That's that's what was thinking because, like, it feels like they're ahead, but like there are other tools as well. Like we talk, what was the one
Bart:A lot of tools like like Google has their own.
Murilo:Google has their own, but like people everyone thinks of Lovable. Like I saw a few people that built a few different things on different tools and I didn't get the clear, like, oh, yeah, Lovable is much better.
Bart:I think that you could say that UI wise, it looks a bit more fancy from the get go. I think you could say that. Yeah. But I hope for them that they have a like a like their their tech is defensible enough for the long run.
Murilo:Yeah. That's the thing. Yeah. Because like making stuff look a bit more sleek, can probably
Bart:I have the feeling that what what they are doing now will become a commodity before we realized.
Murilo:Yeah. I think so. I think so. And then next, we have Alphabet. So Google building chips.
Murilo:Well, actually, they were already building chips. Right? They were building TPUs, but they're trying to basically get a bit ahead by extending the support of TPUs for PyTorch. Yeah. So PyTorch should like, I think it's it's uses CUDA.
Murilo:Right? So there if you're using PyTorch, you put towards a framework for building learning models, so neural networks, including LEMS. And if you're building with PyTorch, you kinda had to use CUDA. Like, it was very easy on the API. It's just like dot two parenthesis CUDA.
Murilo:But now Google is actually working with Meta to bring the support of TPUs
Bart:to PyTorch. Support base.
Murilo:Exactly. Exactly. So it it can be big. I think it's nice. I mean, again, why only now?
Murilo:I feel like this this is something that could have been done a while ago. Right? But I think it could be it could it could increase the uptake of of TPUs for sure. For sure.
Bart:Yeah. So if that's, like like, that's another question. And whereas NVIDIA five years from now, I think that is the the the not a million dollar question and the the how many billion dollar question?
Murilo:Many billions.
Bart:Yeah. Yeah. Because you do see a lot of no one is really like, you see a lot of initiatives that are trying to build performance chips, and and Alphabet with the two b users is one example. But Amazon has ruined a lot of Chinese players these days. And
Murilo:Actually, I think, like, the the them stopping selling chips in China, it's really bad for NVIDIA. Like, in the
Bart:It's really bad for NVIDIA because it, like, it basically forces them to to start building their own chips. Exactly. I think there's even there's rumors now that that they're if acquired well, not acquired, stolen. Quote, no. I'm not gonna say stolen.
Bart:Somehow mimicked the technology of, ASML. Ex employees of ASML have somehow mimicked, sort of simulated the technology to to build this, basically, this base material for chips in China. So also there it is. There's more competition. So, yeah, let's see where, if it gets.
Murilo:And what is EU in this split? US, China, Europe is
Bart:US ASML.
Murilo:So that's also okay.
Bart:The fry factor.
Murilo:Yeah. Indeed. And last but not least, what we have?
Bart:McKinsey eyes thousands of job cuts.
Murilo:And this is AI related as well.
Bart:Oh, that's what everything thinks. Right? Consulting is under a lot of pressure, last year. I think a lot of it this is probably because there's a big AI hype. The AI hype means that all big customers are saying, maybe we should wait for investment because AI will probably solve it in the near run for us.
Bart:So let me Yeah. We wait a little bit. Did you see this effect? Just people being cautious with spending money? That has an effect on consulting.
Bart:But I do think that what is happening also, but not already at the scale that we hear the hype is that things are getting automated. Things are getting displaced. Things are and it's so it's so small, but it's it's it adds up over time.
Murilo:Yeah. Yeah. I also think yeah. We saw also, like, for the the the big consultancy as well, how people are asking for text. No, sorry.
Murilo:AI break. Right. AI reduction or AI. How do say discount?
Bart:Yeah. Like if if you're my consultant, if you can do this, like if 25% more more efficient because you use AI, then I need to see some of that.
Murilo:Exactly.
Bart:Game.
Murilo:Exactly. So I think there was already some some question. We also saw the reports from Deloitte, right, that they were using AI and there was some slop in the middle there, which again is normal. Raises a bit like, okay, if you're just using AI to create all these things, then maybe I should just use AI myself. Yeah.
Murilo:Right? Like, there's no more secret there.
Bart:I see.
Murilo:So so yeah.
Bart:And maybe also to like because it's important detail, like for McKinsey, at least specifically, like they are initially eyeing non customer facing jobs. Okay. So you can also see, like, this is maybe, like, the quote unquote AI effect on
Murilo:Yeah. Like, they're more efficient, they need less internal goals for this.
Bart:But we don't know. Right? Like, what the real reason is. Alright. That's it for today.
Bart:That's it for this week and actually also for next week. For 2025? That's this is the last one of 2025.
Murilo:Yes. We're saying goodbye to 2025. Maybe let me just do some work here. Any any words of wisdom you wanna say before we
Bart:No. Maybe. When did we do the very first episode this year?
Murilo:The very first episode.
Bart:To say April, but I'm not not
Murilo:Let's see. Yeah. I can go here. The very first episode.
Bart:Next page already. Oh, you already can click next page.
Murilo:Yeah. So many. Our first episodes, May. Okay. May 6.
Bart:May 6. We're now, I want to say, 25 episodes in, something like that. Right?
Murilo:Sometime then. Yeah. Twenty eighth.
Bart:Ah, twenty eighth. And we're already have an audience. Exactly. We do. Right?
Bart:We went over 4,000 subscribers on YouTube.
Murilo:Exactly.
Bart:That is cool.
Murilo:That is really cool.
Bart:Roughly, these weeks, like, we easily cross 10 k views and listens across across channels. Yeah.
Murilo:Yeah. Indeed. It's really I think, yeah, it's been really cool to see. Yeah. Cool.
Murilo:Big big uptake. I think, also, we had some interesting guests. I think
Bart:And we started with the interviews. We already have some interesting IDs for interviews for next year. It's actually the interviews were a lot of fun. Right?
Murilo:I think so.
Bart:It's bit, finding the time to fit everything in.
Murilo:Yeah. Yeah. No, for sure. I think also we, we went from, from podcast to YouTube, TikTok, Instagram, newsletter. So I feel like we've done quite a lot of stuff, and I think we also found a nice temple that it doesn't like, I think the only reason why we're able to do these things is because it's not like it takes such a toll.
Murilo:Right? I feel like we can we can carry on our day to day. So which I think is also important.
Bart:Yeah. True.
Murilo:True. Right? Finding a balance, like, okay. We're gonna edit these things, but maybe we should spend more time here, but maybe that you know, like, okay. Maybe can this be more efficient?
Murilo:I think it's also it's also important.
Bart:Yeah. Now we have some we have some IDs, but maybe that's a bit early. I want to say we have some IDs for an actual event next year.
Murilo:Yes. We do. That is true. Maybe maybe it's a bit too early, but maybe we
Bart:can say what we're thinking about without promising it.
Murilo:I think I think that's a good Christmas gift.
Bart:That's good to put to put a bit of pressure on ourselves as well.
Murilo:Sure. Keep us accountable. Yeah.
Bart:So what do we want to do?
Murilo:You, please.
Bart:So the idea is that, we are here locally at least, but we will record everything Yeah. And, like, share it like like we usually do with a long format episodes and shorts, but we want to do a local event for either young young startups or aspiring founders to pitch their IDs and basically bring these this entrepreneurial and investment community together
Murilo:Exactly.
Bart:For very early stage companies.
Murilo:So people that are like, they wanna they have an idea. They want to to to hear thoughts or maybe, like, just to share some cool stuff or get people excited or, like, have a bit of a platform where they can Yeah. They can they can get more visibility, get a bit of the spotlight there. Right? So
Bart:so And we'll try to also have this idea with that we have people that actually like, the audience can also be part of the jury, but we need to see, like, how we make it that we've tried
Murilo:Yeah.
Bart:Find the right people for that. Right? Yeah. Yeah. Yeah.
Bart:That easy. Yeah. And I also like to make it very easy to network. Right? Like like Yeah.
Bart:Simple as, like, my name like my name sticked on my on my T shirt or or, like, maybe the the my name in a certain color. If I'm looking for a co founder or something like, we're thinking Like about this
Murilo:you're looking for someone to help you to, to go on these things. You have a nice idea, but maybe you're looking for help or maybe you're looking for help from like, A lot of things there, but stay tuned. And, yeah, I think that's a, that's a rep for 2025.
Bart:Yeah. And now with all the money that we made from the podcast, we can go out for dinner. We're going
Murilo:to have like a water. No. But, yeah. Thanks everyone for for following us this year. Yeah.
Bart:Thanks a lot. It was, it was a great journey.
Murilo:Yeah. For sure. Thank you, Bart as well.
Bart:Thank you, Michaela.
Murilo:Merry Christmas to everyone, and happy New Year. We'll see each other in early early next month.
Bart:See you in the next year.
Murilo:Ciao.
Creators and Guests
