AI’s Wild Week: Apple Sued, Anthropic Pays $1.5B, Mistral Raises €1.7B — Plus Atlassian Buys Arc Maker & Azure Hit through Red Sea Cables
Hi,
Murilo:everyone. Welcome to the monkey patching podcast where we go bananas about all things lawsuits, hallucinations, and more. My name is Murillo, joined by my good friend Bart. Hey, Bart. Hey, Murillo.
Murilo:How are you?
Bart:I'm doing very well.
Murilo:Very well. New mic? New mic. Yeah.
Bart:Yeah. But I'm gonna setup.
Murilo:Yeah. Che. Wow. Like, you're holding, a trophy. Like, you're very proud.
Murilo:Like, you're like Yeah.
Bart:The stand looks a bit like a trophy. Right?
Murilo:It looks like a it's like a podcast trophy. Why'd you get a new like, you wanna fill people in? Like, why'd you get a new mic?
Bart:I had a lot of issues with my last one. The sound was a bit distorted. Is that think it's actually the it's not the mic itself, although it might be. My kid dropped it once. Might be.
Bart:But it's like the it's an amplifier. You need to have an amplifier in between your mic and and your computer. Don't think it's a problem with the amplifier. It plays an amplifier. It's not really meant for mics.
Bart:It's also meant for guitars.
Murilo:Do you play the guitars, Bart? Oh, you I you diddle. Okay. Few years back,
Bart:I sometimes diddled the guitar.
Murilo:Yeah. What's your do you have a favorite song that you like to play?
Bart:No. Not really. Not really. I just try to play something. It's a slide guitar, actually.
Murilo:Slide guitar?
Bart:Yeah. Something that you like, you put it flat on you on your lap.
Murilo:Oh, really? So is it
Bart:lap steel guitar.
Murilo:Wow. Didn't know that. Okay. Very cool. So next episode, you prepare the intro jingle.
Murilo:I
Bart:would just wanna do that to people.
Murilo:Okay. It's like listeners drops. Right? Like, alrighty. Cool.
Murilo:What do we have for this week? We have quite a lot of stuff. We have some tidbits at the end as well.
Bart:Yes. I'll I'll kick them off.
Murilo:Kick them off. Kick us off, Bart.
Bart:Is buying the browser company for 610,000,000 to build an AI first browser tailored for work. The plan centers on Dia, the startup's new AI browser with ARC continuing but taking a back seat. If Atlassian nails this, fewer app switches and deeper Jira and Confluence integrations could make the browser feel like your workspace.
Murilo:Alright. So another browser company being well, another move in the browser game?
Bart:Yeah. Exactly. For a lot of money. Have the feeling like, like, since, since this whole AI AI race started, like, no one is is batting an eye at these numbers anymore, but it's still absurd.
Murilo:Yeah. It really is. It's it's the browser company, which is actually from, like, 2000 yeah. '19.
Bart:Yeah. I think the browser company, I think why it was most known to me is because of the ARC browser, which a bit predates the whole Geniei stuff. Right? Yeah. It was a bit of a like, a browser that's was a bit anti clutter Yeah.
Bart:Trying to make you more more more, let's say, performant. Yeah. More efficient workflow in your browser.
Murilo:It did get some
Bart:recent years, they also introduced Dia, which was like an AI powered or is an AI powered browser where you can basically interact via text with your the tabs that you have open.
Murilo:Art got some traction there, right, for a while. I remember was very trendy, very hip.
Bart:I think in the tech community, it was very hip. Yeah.
Murilo:Not sure
Bart:if outside of that, people even knew it existed, but I think in the tech community, it was very hip. Yeah.
Murilo:Yeah. Indeed. Do you know what the and also, Dia, what does it stand for? Do you know?
Bart:No. I don't know.
Murilo:Did you know that Dia in Portuguese is day? True story.
Bart:Interesting. Indeed. Indeed.
Murilo:Lot of money. Yeah. They say those.
Bart:What Atlassian will do with it. Right? Because Atlassian, like, they're very well known for, like, their, let's say, developer work suite or developer workspace, that kind of tools, Jira, Confluence, etcetera. Yep. Bitbucket, also a big one.
Bart:They now have a browser. It is, of course, interesting. Like, if you integrate all this very tightly within your browser, like, your browser can really become your central workstation knowing that a lot of, like, even development environments, like, they run very well in browser these days. So maybe Atlassian is basically trying to pull everything in and have, a central enter entry point, which also makes it a bit easy or more more difficult for people to leave because, like, they have everything out of the box.
Murilo:It does feel a bit out of left, like, out of nowhere, right, for Atlassian in the sense that when you mention the tools that they have, everything feels like it kinda comes together, but then a browser feels a bit I mean, I'm curious. I feel I'm sure they have a a strategy for this, but to me, it's not clear yet. It's not like a clear move. Right?
Bart:To me, how how I how I would see it, like, they they have these developers slash project management tools. Right? And by adding this browser, and you can run a lot on the browser these days. And if you integrate all this, like, it becomes a bit like your operating system for your workspace.
Murilo:Yeah. True.
Bart:True. And maybe that's a bit what they're trying to do. We have this this central entry point where it's we also becomes hard to leave. Right? Because everything is there, and it just makes sense to stay.
Murilo:True. True. True. Yeah. For me, I'm makes sense, but I'm also wondering if there are other ways to do this.
Murilo:Right? Like, an extension in Chrome or doing this or doing that. Right? Like, you do need to
Bart:Well, there there probably are, but I can also imagine, like like, their customers are typically the large enterprise of this world, right, where it's always hard to do device management. Like and if you have your your whole development experience within your browser, it also becomes easier to, let's say, enforce enterprise controls, like, more fine grained controls about things you access, whatnot, etcetera, these sort of things.
Murilo:Yeah. That's true.
Bart:Actually, it's a hard thing to pull off, but I think it makes sense if you think in that direction.
Murilo:Yeah. That's true. That's true. I mean, I think I'm for sure I mean, it's a lot of money. Yeah.
Murilo:So for sure, they they had a
Bart:don't know what the, like, the migration path for existing ARC users will be. Like, you're you're using your browser suddenly. Like, you get a pop up This to tap now turned into your Jira issues overview. Exactly. Yeah.
Murilo:Yeah. That's true. That's true. It's like every ARC user is gonna be creating Jira tickets now, you know, for personal life. Yeah.
Murilo:Do you know the numbers? How many people were using ARC? Because the feeling I get is, like, it got a bit trendy, but then it also died down a bit.
Bart:I don't think it is a lot. And because I in a in a lot of context, I was looking at a benchmark the other day, and I don't remember that I even saw ARC in in browser usage. But so it's probably not much, like, points, probably.
Murilo:Well, I think browser nowadays nowadays, the browser game is very difficult. No? I think Firefox is and we talked about it some maybe a couple years ago, maybe already, that today is, like, ninety nine percent. No. Not 99, but, like, high high high majority is Chrome.
Murilo:Right? Are you pulling do have these numbers?
Bart:No. I don't have them.
Murilo:K. So, yeah, let's see. Let's see. A lot of money again. I'm sure they have a plan.
Murilo:Curious how how it's gonna unfold. True.
Bart:Yeah. Let's see.
Murilo:Shall we move to our next topic?
Bart:Let's do that.
Murilo:Let's do it.
Bart:We have Mistro case of big numbers.
Murilo:Yes. Well, we have a lot of big numbers today. So we have Mistral AI says that it has raised 1,700,000,000.0, a huge shot of capital to accelerate its technological push. The announcement signals bigger ambitions for model development, compute, and go to market across Europe and beyond. For customers, expect faster releases and another strong alternative to the usual US centric AI platforms.
Murilo:I think they had, like, the last so Mister AI, I think, is the big EU player, right, when it comes to large language models. They have the LeChat, which is like the LeChat. The GPT.
Bart:Alternative. Right.
Murilo:Alternative. They do release way more open source models, which is already very different from what OpenAI has been doing. And I think they had the the last round of funding was last year, I wanna say. And now they had another round of funding funding, and they had 1,700,000,000.0 there. And I think the biggest one of the big, big investors was ASML.
Murilo:Right? And on the LinkedIn post, which I have on the screen, you also see here. Industrial partnership with ASML is to support the global semiconductor equipment manufacturer to keep pushing frontier AI to solve the most critical and sophisticated technological challenges faced by highly complex industries. And I think with that, ASML also has a board seat.
Bart:Yeah. I think ASML actually becomes the the top shareholder. They will hold around 11% of of shares.
Murilo:Indeed. Indeed. Were you surprised by this part?
Bart:Surprised this a lot to say. A little bit neutral. I think it's interesting to see. I think it's interesting to see that but I think what this about this round this funding round shows is that big investors I mean, there are big big names here. They are hopeful about Mistral AI.
Bart:So I think that's that's maybe the way to to phrase it. I think that's a positive thing. I think maybe I was surprised about ASML leading around. Not sure what that will mean towards the future. ASML is much closer to the chip production side, to the hardware side.
Bart:Right? They don't supply chips directly, of course, but they are much closer to hardware. Maybe we do see some European alliance between AI, hardware, and software going forward. That's interesting to think about, I think. At the other side, I think in terms of benchmarks, in terms of community, Mistral is still very small when you compare this to the rest of the world even with, like, a huge valuation.
Bart:Right? Think they're valued at 11,700,000,000.0 post money. But at the same time, they are doing some revenue. I think the the latest reports coming out of this is that they will this year, they would expect to do a 100,000,000 in revenue, which is, of course, penis versus their valuation, but it's still a 100,000,000 in revenue. I think what they are doing because there was also a and I think goes a bit behind the the like, it's not really visible to the to the community, but they are focusing very much on enterprises.
Bart:They had an announcements last week that they are integrating very tightly with 20 very big enterprise And that's maybe also the the approach that they're taking because, of course, they are the only serious European contenders, so it makes a lot of sense to focus on European enterprises to really have a top notch offering for them. And, like, this funding round and who's leading it, like, maybe also shows that they do have traction in this in this space. Right?
Murilo:True. True. True. True. And, like, to say to their leading, it's also because it's 1,700,000,000.0, and I'm looking here at this the ASML announcement.
Murilo:1,300,000,000.0. What was ASML? So it's, like, really, like, almost just ASML, right, investing.
Bart:The I think it's ASML. I think a lot of the ones that are mentioned as well, they were I'm not sure if it's all, but a lot of them were existing investors that basically followed around. And, in fact, they they they increased their their previous investment.
Murilo:Indeed. Yeah. Again, really big numbers. I think it's good that people are investing in the EU's
Bart:AI. And I think it's good I think it's a good strategy that they're following, like, to really focus on this this enterprise enterprise layer within the European Union, which is still very big. Right? Like like, let's be honest. Like, it's in in terms of global market, like, the European Union is almost as big as The United States, right, in terms of people that are there.
Bart:Maybe it's a bit small in terms of total enterprise value, but, like, it's still very, very significant. And Yeah.
Murilo:It sure.
Bart:Make sense to to try to compete with the global market, but you should just compete within Europe.
Murilo:I think also Europe has more I mean, to share the European Union values, right, like, on the the privacy and all these things, I think it's is a differentiator. Right? I think there's a lot of Especially with current geopolitics. Exactly. There's a lot of fears, maybe quote I don't know if fears is the right word.
Murilo:Right? But, like, do we want to rely or overrely on opening eyes and all, you know, on tropics of the world? So and I think Mistral, I mean, at least from my perspective, right, you don't see as much you don't see as much controversy as you see for anthropic. We're gonna talk about anthropic later. I mean, also, open AI, right, on a lot of these things.
Murilo:So maybe it's also because it's smaller. To be very honest, you don't hear as much in general, like the good and the bad, but I'm not sure if this also the the the care, right, for for human values and all these things. I don't know if it also permeates on their models. But I think it's good in any case. I think it's important.
Bart:What else do we have? We have here Alex Kantrowitz argues that generative AI now has three jobs, getting things done, developing thoughts, and companionship. He cites GPT five's agentic bend while deeper reasoning comes with cost trade offs. For users, it's a reminder to pick tools by job, execution, exploration, or emotional support. What is this about, Milo?
Murilo:Yes. So it's an article that I see passing by. It resonated to me, and there's also a bit of a controversial use. So what he's arguing in this article is that now it's been almost three years since SEGPT came out, and I the industry has a bit Uh-huh. Of
Bart:It's been.
Murilo:Crazy. Yeah. It's crazy. But now the industry is a bit converging on different uses of generative AI. And I think he mentions agent, which I'll put a bit of an asterisk, but agent to his in his definition is just to get stuff done.
Murilo:Then he mentions thought partner, which basically means chatting about things, discussing. When I first read thought partner, I thought of sometimes what I do, for example, I say, hey. I have this plan. Can you criticize my plan or ask me follow-up questions? And sometimes the questions are like I mean, a lot of times I'm writing a document, so I want to be thorough.
Murilo:Most of the times the questions are like, okay, just forgot to type this. But sometimes every once in a while, was like, oh, actually that's a good point. Maybe I shouldn't maybe I should think it through. Right? That's what I thought he was he meant, and it's kind of what he meant, but he's also talking about like reasoning models.
Murilo:So something that is really it's not necessarily about the task, it's more about the output. Right? And the reason why I said about an asterisk about agent and thought partner is because the thought partner, according to him, can also be agentic in the sense that he can also make tool calls for the web and like
Bart:Okay.
Murilo:You know, reason and all these things. So I think the main difference is
Bart:becomes a bit like, yeah, your deep research mode or something.
Murilo:Something like that. Yeah. But it's not just to get stuff done, and that's how he defines it. Right? Like, top partner is really like, what is the what is the outcome here?
Murilo:It's more about the conversation, not about the outcome. He there's a quote here. Right? Like, the agent is searching for efficiency and wants to move on, move you on to the next thing, the thought partner is happy to dwell. Right, and then actually meets you and make sure that you understand things.
Murilo:And then the third, well maybe before we go to the third one which is a bit more controversial, Does this also mirrors how you use AI in your day to day?
Bart:Yeah. That's that's a good question. I have a hard time putting this framework on how I use it, to be honest.
Murilo:Yeah. But that's because it doesn't fit? Or
Bart:I have a hard time fitting it. Yeah. Like, to me, they're like, like, I have some automation tasks. And then I have I use it also quite a bit in, like, discovery slash learning mode, like, to understand, like, these and these new articles to give me a summary of these articles and to really drill down on articles, for example. Yeah.
Bart:I do this when we prepare for when we prepare for the podcast. It really helps. I think that's it would fit here maybe already in thought partner, but to me, it's also, like, a bit to get things done. The thought partner, like, I I would fit a bit more than the deep research stuff on there then maybe
Murilo:Yeah. I see.
Bart:Where you if a bit more complex question,
Murilo:like,
Bart:I have this set of companies, do some research on these and these man metrics, give me a report in the style. Like, maybe that's
Murilo:Yeah. It's a bit murky. Right? Because even that, you could argue that you're just getting things done. Right?
Murilo:Like, you're Yeah.
Bart:It's it's to me,
Murilo:a bit delegating to the yeah.
Bart:And also, like, to me, like, the I also do some AI assist coding that would probably be an agent then. I do some there are also creative parts. Right? We've we've discussed music. We've discussed image.
Bart:I think it's also I have a hard time fitting it here. So I'm not sure. Like, it's I wouldn't use this for Maya.
Murilo:Indeed. Indeed. I also don't I read it, I thought it somewhat resonated, I guess. I also don't I wasn't convinced that it encapsulated all the uses, and that there's a third category that I'm leaving for the last, which is companion. I think, actually, between the three, this is the most well defined, let's say.
Murilo:And this is what he says also the most controversial. And he said perhaps the most popular, maybe, and I haven't checked this. But
Bart:Well, this
Murilo:is We also saw I think last week, I think we were seeing, like, the the uses of the Gen AI, and I think the third one was the Second one. Second one.
Bart:Second one in usage has came from the the Cloudflare AI insights report. And the second biggest AI crawler, I think, was Character AI. Wow. Character AI is basically like you can build your your companion.
Murilo:Yeah. And, also,
Bart:it's a it is very, very big in China. And it's probably bigger here in Belgium or in Europe. It's probably bigger than we than we realize also, right, if you look at these numbers from Cloudflare.
Murilo:Yeah. I'm also wondering how I don't know. I don't know if it's also something that behind the scenes people use it, but they don't advertise as much. Right? Like
Bart:Exactly. Yeah. I agree. I understand what I mean. Yeah.
Bart:Like, you're a bit hesitant to talk about this because it's still a little bit frowned upon. Like, Like, you're Yeah.
Murilo:I mean, it is also controversial. Right?
Bart:Maybe is a bit weird, like, AI companion. Like
Murilo:In this article, just to they also link to this post, I guess, which is, like, how to use people generative AI in twenty twenty five. Apparently, it's from Harvard, so haven't checked this. But according to them, in 2025, the first thing is therapy and companionship Wow. I guess. Yeah.
Murilo:It's crazy. So yeah. So back to
Bart:the article. What is the top three? Maybe the Top
Murilo:three from this Therapy and companionship, organized life, find purpose.
Bart:Oh, wow. Yep. Very good. Right?
Murilo:Find purpose. Yeah. Indeed. Organized life, maybe that's more like the ad hoc things. Right?
Murilo:Like
Bart:I'm I'm surprised the generate code is, like, the it's, fifth. Like, it it shows how much in a bubble that we live. Right?
Murilo:Yeah. Indeed. Indeed. Yeah. Like yeah.
Murilo:True. True. Enhance learning, what you said here as well, number four, which actually I think this makes a lot of sense. Even though I wonder the people that use it to enhance learning, how how are they using this. Right?
Bart:Yeah. You need to you need to learn a bit how to learn with this.
Murilo:Exactly. It's a new skill. Right? But it can be very, really
Bart:It's hard for me to explain now because, like, I use this a lot in the car, like, when I'm on hands free, the the voice mode on JCPT. And, like, you sometimes you hear maybe this is not really complete correct. And then you need to ask, please verify this by crawling relevant sources and then cross relevant source. But you need to do know when to do it. Right?
Bart:Yeah. Yeah. You know when you need to you should question the results.
Murilo:Yeah. That's true. That's true. I think yeah. Indeed.
Murilo:But it's kinda like it's like you have a friend that he's very confident all the time, but sometimes he just spits shit. Like, you have to be like, are you sure? You know? Like Yeah. Yeah.
Murilo:But yeah. Indeed. Back to the article. So he mentions here, most controversial. Right?
Murilo:Maybe most popular is to use this as a friend or lover. And he also mentions a few stories. Right? So I think not that long ago, there was OpenEye got actually sued because there was a teenager that took his own life, actually, after chatting with Chad GPT.
Bart:Sad
Murilo:story. Yeah. Indeed. Very sad. And, apparently, like, through all the conversation, again, it was the the how do you say it?
Murilo:Syn syncom psychopathy? How do you say it again? Fancy. Psychofancy? Are you just reading the South Park episode title?
Murilo:Psychofancy, basically, like, the the tendency for the model to agree, right, with user, basically. Right? And, apparently, the the the teenager was suicidal, and he would confide a lot to CHJPT. And then let's say it it enforced or empowered. Right?
Murilo:Even it asked, like, for for tips on, like, should I do this? Should I do that? Would this work? And then was very, like yeah. Like, instead of trying to be a bit more safe.
Murilo:Right? So OpenAI did receive a lawsuit in the past. I think in the past, we also discussed AI as a therapy. Right? And I think back then also, we had also met I'm I remember saying that it can help a lot of people because there's a bit of stigma around therapy.
Murilo:Right? And I think having something very accessible, but at the same time, there are some cases that are very serious.
Bart:Slippery slogan.
Murilo:Exactly. Right? So when it's bad, it can be very bad. Right? And he also mentioned some other cases where it's not so disturbing.
Murilo:And, apparently, AI robot dolls charm their way into nursing the elderly, and GHBT power robot seven year olds are taking over some work from caregivers to lie to seniors to treat them like grandchildren. And I think this is South Korea as well. So another use of GenAI to simulate a companion, I guess, and this is for
Bart:It's also a bit worrisome. Right?
Murilo:It is. It is a bit I don't know. It's a bit, like, You know? Right? Yeah.
Murilo:So, again
Bart:Like, it's one thing to, like, get a dog at old age. Like, it's cuddle and something to take care of. It feels a bit weird to have, so, like, something that that tries to mimic a seven year old and says, grandma, can I get a kiss?
Murilo:Yeah. Exactly. It's a myth.
Bart:I mean, not sure if that's good thinking we should be doing it as a human race.
Murilo:Yeah. Like yeah. Yeah. I don't know. I mean, I have the same impulse as you, but then at the same time, if it helps, if it's you know?
Murilo:Like, is it just weirdness from me because I'm, quote, unquote, old fashioned? I'm not sure. But I I have the same
Bart:To me, the the most worrying thing is this these are things like Character AI where you build, like, artificial friendships, artificial romance. And I think you hear a lot of sources already saying, like, the younger generation has a lot of trouble making valuable connections in real life, this will not make it better on the on the contrary. And, like, you you get these these these short lived things where you get some some validation, like, endorse someone that wants to interact with me. Yeah. It becomes easier not not even to think about the next step.
Bart:Like, I I actually need to go out and make actual friends anymore.
Murilo:Yeah. I agree.
Bart:I don't think this is a good thing for for our social life.
Murilo:Yeah. I think it's one of those things that, like, we're getting so comfortable as species like society. Right? That if you just lean it lean it too much, it's actually not beneficial for you. Right?
Murilo:I think a level of challenge and hardship is good in your life. And you also see, like, I think in Japan, they have like, the loneliness is actually a big problem. Maybe I mean, we saw that the other thing was from South Korea. Right? But
Bart:But that's a bit what I mean. Like, if loneliness is a problem, this feels like
Murilo:It's a bed date. Right? Like, you're treating the symptom, not the disease.
Bart:Yeah. But you start using this, and maybe it it indeed feels a little bit less lonely. But it's it's you treat the symptom, and thus, you don't even start to think about about more, like, do we make certain
Murilo:Indeed. Indeed. I mean, I feel like, again, it's like a it's like a band aid. Right? Like, it helps the problem for today, but the problem is still there.
Bart:Exactly. Yeah.
Murilo:Right? And the problem is not gonna go away. In fact, it makes it even worse. I think there's also the whole if you find if you create a companion that always agree with you, I think it's also easier to get to the extremism that we see in the world sometimes. Right?
Murilo:Like, always like to. So I think, again, being being challenged sometimes is good. Right? So he also mentions this. I think, again, it's something that we don't talk we don't cover as much, I feel, but, apparently, there's a big group of people.
Murilo:And one thing that I wanted to share as well, like he says, when you're building an eye product, part of the trouble is some people always fall in love with it. And I think it means fall in love in the very romantic sense. Right? And he says, yes. There's even an erotic fan fiction about Clippy.
Murilo:And I just wanted to share that. You know? Like, pray this conquered by Clippy, an erotic short story in Goodreads. And then there is one review. There's like, I'm only admitting to reading this because I'm behind on my reading goal.
Murilo:So yeah.
Bart:So yeah. Yeah. Tomorrow, you're
Murilo:It could be worse.
Bart:Creating a Character AI account, and you're you're not creating a Clippy.
Murilo:Yeah. Exactly. It could it could always be worse. That's, I guess, a that's what that's what I that's what I wanted to say. Alrighty.
Bart:Clippy band. Sorry. Let's go to the next one. What do we next,
Murilo:Move on quickly. Simon I don't know how to say his last name. Spati makes the case that relying on AI for core writing or coding dulls skills and produces bland work. He even turned off Grammarly and Copilot to regain focus using AI only for quick autocomplete or diagrams. If you build or create, his advice is simple.
Murilo:Use AI deliberately. Keep your mental muscles strong. I read this article as well. I think it's not very unlike other things that we have also seen. There's also research, right, about creative writing from MIT, I wanna say.
Bart:And it's a bit of a like, the article, like, it combines a lot of opinions of other people. Right? It's it's it's and also some talking about frameworks, like, bit how can you think about where it can help, where it does not help, where where the what is productive or is not productive. It's and a little bit tongue in cheek as well. Yeah.
Bart:But it's an interesting discussion on, like, what does it do if you relied too much on this? Like, does it only improve your productivity, or do you actually start losing skills? I think that's a bit of the premise that has been created here.
Murilo:Yeah. There's also some, he doesn't say technical debt, but he also says some things that makes me think of technical debt. There is the muscle part, which I think it's true nonetheless, but I think some of the arguments against the use of AI made me feel like he's saying, like, using AI and just, like, letting it rip it, you know, like, without supervision. And I think for a lot of the things, for example, one of the things he mentions is, like, architectural designs are big no no's, which I think it's true if you're if you're letting the AI have full agency. Right?
Murilo:But I feel like if you're if you have someone there, like, that is verifying, that has their opinion, that is they're, you know, like, an expert, which again, is it is it Genie I mean, maybe it's Geniei assisted, maybe get some ideas, but I think there's a bit more nuance as well for some of these things. Right? So, yeah, he has some interesting diagrams here. Like, one is the he he sources a lot from other sort like, sources a lot of quotes and and images. Right?
Murilo:And I think this one was also also funny. Like, the illusion of productivity, of AI productivity. Right? So as time goes over, you're, like, you skyrocket, and then you you tweak, and then you go back down. You have to restart, and then you go back, and it goes a bit higher, and then you okay.
Murilo:And then you go back down. And you keep this pattern until it's good enough, and, actually, you you you plateau. And, yeah, again, the images has also, like, a before AI that was, like, steadily semi steadily going upwards. Right? Again, not sure if I agree with all the things.
Murilo:I do think that the muscle part, I think, is the most relevant in my opinion. Right? And he also there's another part of, like, writing or not writing that he says, like, writing is also a like to thinking. So if you have people that are not writing, you're actually taking steps towards people that are not thinking. Yeah.
Murilo:It's
Bart:it's also slippery slope. Right? Like, I was I was actually thinking about this the other day. And do you have this book called thinking fast and slow or something?
Murilo:Yeah. Yeah. It's a popular one.
Bart:I'm not sure if that if the title exactly correct. And I was actually thinking about this that's the the concept in this book when it comes to my own use of Chen AI. And, like, the concept of the book is like you have two systems. Right? Like, two modes of operating.
Bart:Like, you have a system on where it's, like, it's a bit instinctual, like, very quick very quick decisions. Also, very biased. Like, you think you very know very well, but it's very automated, and sometimes you make mistakes. And system two is a very much deliberate like, it takes time and takes effort. And in real life, it's not one system or the other.
Bart:It's a combination of these things. But I was thinking the other day, like, my own usage on the on the AI when for example, when it comes to learning. Right? Like, when it comes to learning a completely new domain, whether it be in tech or business or whatever, like, it used to take a lot of time and effort to find targeted information and to study that information and to do something with it. And now you ask Chechipati and they think, oh, okay.
Bart:Now I know. And then you move on to the next thing. Like Yeah. You are very much more in this system one, and you don't really train your system to, like, where you're you're you're not really training. Like, this is a deliberate system that makes your most muscle better to than actually do instinctual stuff with it.
Bart:Right? Like, it's it's so much more like, and also when when when to make maybe the the comparison layer with with with coding, like, when I use a new framework with that I haven't used before and I use it now with GenAI, I study it way less before actually implementing it than I used to do because I used to have to go to the website and look at tutorials, see this is how you use it in these in these scenarios, and then, like, build a bit of a a bit of an architecture in my brain. Like, this is how you typically do this, and these are what best practices. Now I start with the and use that framework, and then it does something for me. And then, hopefully, it's correct, and then I try to figure out along the way.
Bart:But there's a completely different way of picking up new stuff, which I do kind of for myself worry a bit about. And I noticed it as well, like, because you mentioned writing here. I mentioned it. I I noticed it in writing as well. So I have a harder time to these days sitting still to write a large portion of text than I had before JGPD simply because you're so used to generate this text and needs to have this content.
Murilo:You know what saying? I think for me, I rest I relate to some of these things. I think but I think before, it was, like, I struggled. Like, I have a hard time for some tasks, and I think, again, I think I have some ADD traits, and I have this hard time sometimes just sitting my ass and just, like, I'm gonna read this. I'm going to understand this.
Murilo:I need to understand that. And I feel like with AI or GenAI, there's there's some shortcuts you can take now. Right? And I think I still do the work to to when I feel like I I shouldn't be taking shortcuts, I still try to do the work, but I also noticed that I'm way more impatient in a lot of ways. Right?
Murilo:Like, I'm like, I'm doing this now, but I'm like, fuck. This is taking so long. You know?
Bart:Yeah. Yeah. Like, I should
Murilo:be doing this. I should be doing that.
Bart:Oh, that's what a bit what I mean with the writing stuff. Right? Yeah. Indeed. No longer use that to it.
Bart:That writing stuff takes time.
Murilo:But do you think do you feel like you're you're maybe losing that skill, or do you think it's more that you're impatient?
Bart:Maybe a bit of a combination. Like, maybe you become impatient because you're able to create stuff so quickly or get an answer so quickly with something like JGPT. And on the other side, like, maybe losing, not necessarily specific writing skill, but more like a deliberate thought process skill to really in-depth struck in a structural way, go to something, research something, and, like, build knowledge in that field?
Murilo:Yeah. I think yes. But at the same time, I do for example, I was studying I had to study like a set of eight documents. Mhmm. And then the documents, they were just they just said, these are the documents.
Murilo:And they all had like different like, they didn't give the context of what was the document about. So some documents are really well structured, some were not, but it also when I was trying to read, I also saw that there was some disagreement, let's say. So I was like, okay. Maybe this is outdated. And I actually used Notebook LM, not to replace.
Murilo:It's a Google product. Yeah. So for people that don't know, like, it's a now it's a study companion. They got very popular because it would create podcasts of the source material that you put. So I used it first to, like, okay.
Murilo:I wanted to to listen to it first, like, bit of a discussion, because sometimes I think it's helpful still. I I feel like I nowadays if I'm reading something, before I'll be like, every line I read, I need to understand it. And sometimes I'll read a paragraph, and then my mind is wandering, and I'll go back to paragraph, read again, read again until and nowadays I'm like, try to get like a skin through first, then try to dive deeper into each one of those. And I think for that like the notebook LM actually helped me in the sense that I could listen to it, and they would talk about some things, and then they also have a bit of a mind map that you can kind of click and actually expand stuff, then you also have a bit of a chat and always with the source. Right?
Murilo:So you can click and zoom in the article. So I feel like in a way like it didn't really replace my my the reading, but it guided a bit more. Right? Like, okay. Now you have this.
Murilo:Now you have that. It it just kinda gave me a bit of a structure. And I think sometimes even in learning, yeah, like, if you have research papers as well, some of the papers are really well written, and some of the papers are really not. And it took me also some time. Like, in the beginning, I always thought it was my fault, but then I I remember I read a paper.
Murilo:Was like, wow. This is, like, super well explained. Right? It's like and and then sometimes, like, okay. Maybe it's not just it's not about the maybe it's not entirely my fault.
Murilo:Right? Maybe the way that the material is presented also makes a big impact. Yeah. And
Bart:that is also where GeniAi makes a big difference. Like, even even if a paper is written very well, but, like, it's a very specific domain that you don't have any knowledge of, like but then use a very technical jargon, then you can ask or test your PD or whatever, like, present this in a bit more layman term to me or generate a concrete example for me that I can understand without having the background, like, and it all really translated to something that is understandable for you. So and then and fully agreed. I get that sense. Like, it's I use it a lot.
Bart:I, like, I very much see the value. But I do sometimes worry about, like, what with these more? Should I not be doing more of these deliberate processes and and and what if I don't do it for a long time?
Murilo:Yeah. That's true. I mean, see, I I I share again, maybe not to the same degree, but I do share a bit the the the the fear, quote unquote. Right? But I'm also wondering, like, my dad, when he was writing his thesis, he had to go to the library, pick up books, go to table of contents, read the stuff, and then cite everything.
Murilo:Right? And maybe when Internet like Google came out, it was like, okay. I mean, he probably didn't. But, like, imagine, right, like, people are like, ah, but I think I'm losing the skill of, like, finding the right book and going to the you know, like That's true. Reading this stuff.
Murilo:And maybe, like, yeah, but maybe you're fine, bro. You know? Like Yeah. That's true. You know, maybe maybe, like, you're learning new skills.
Murilo:Right? So, again, I'm not I'm saying this as well, but I do relate to what you're saying. Well, I mean, again, in general, it's so broad. There's so many tasks. Right?
Murilo:You can be generating music to generating images to listening to whatever, doing your research to writing. So I do think some of these things that we probably should still be doing. Right? I also think that if you leave the AI to do some of these things, it's also probably not a good output. Right?
Murilo:I mean, not good in the sense that one example they gave here, for example, the cover letters. Nowadays, if you recover letters that are AI generated, you can definitely tell that they are AI generated. Right? And I think maybe in the beginning, people were like, cover letters, like, woah. This is amazing.
Murilo:Like, you can read it. It makes sense. But, like, nowadays, like, it's like you get a bit saturated. Right? Like, the same thing with, like, AI generated content on YouTube or whatever.
Murilo:Right?
Bart:And it's growing very quickly. Yeah?
Murilo:Yeah. Because it's it's so easy. It's so easy to do. Right? So, I mean, even a few times, I think you mentioned.
Murilo:Sorry?
Bart:You're still real. Right?
Murilo:Hey. Am I? No. Maybe you
Bart:wasn't at Character AI.
Murilo:Yeah. It's all simulation, Bart. They also mentioned, like, soulless. Right? And I think that's a bit the vibe that you get.
Murilo:Right? You you just gotta do, like, things, like, not really I don't know. It's like like, what's the you know? Like, you get a bit I don't know. There's no substance to it.
Murilo:Right?
Bart:And I think, like, maybe that's a bit the the for me, a bit the the summary of all of this is, like, all of these tools, they are super valuable. I think they for me personally, like, they accelerate what I do a lot, but you need to be remain skeptical about, like, what do you actually achieve with it? Like, is it good we use every any domain? Like, are there reason to say now I need to slow down a bit or really go in-depth on that instead of having a very quick answer and then going to the next thing? And you can still use, like, AI tools to go in-depth.
Bart:Right? But I think you need to know that you need to switch a bit your your own mindset to do that.
Murilo:So I think it maybe links a bit to what we were saying before. I think it's like you need to be consciously choosing maybe the harder path and not as efficient path, right, to because you know that there are benefits to it. Right?
Bart:To the good. Yeah.
Murilo:I think the way I like to and I I mean, maybe I'm taking a bit too be a bit too philosophical here, but I think in my life, I think in general, what I think the way I put it is to be intentional. Right? Like, you're doing things because okay. This, I'm just trying to get it done, so this is fine. This is something that I actually want to learn, so I need to do the work.
Murilo:Right? And whatever you're doing, just make sure you know why you're doing these things. You're not just leaning on the easy the easy path always. Yeah. Right?
Murilo:I just wanted to do a quick shout out here. So he's actually asking, will will he replace x and one is data engineers? And he says, probably not. Nice comparisons by Mehdi Mehdi Oase.
Bart:Yeah. You know, we had him on our previous podcast. We had him
Murilo:on Exactly.
Bart:That's right.
Murilo:We should we should have him on because he's moving back to Belgium, and we should have him on this one. The friend friend of ours. So he says nice comparison by Mehdi. Did the music record replace musicians a hundred years ago? Nope.
Murilo:We just changed them in the industry. Did cloud and computing take all IT jobs? No. We just changed the industry in our jobs. Same here.
Murilo:It would change our industry and the job, but we won't disappear. So I also agree with this statement from Eddie. Right? I think it would change a lot of stuff, but I don't think it will replace necessarily.
Bart:I don't think it will replace software engineers. I agree. Maybe it will replace subsets in that.
Murilo:Maybe. Or change. Yeah. Yeah. Indeed.
Murilo:But I also think that if you're not using, then you should definitely be using these things. Right?
Bart:Let's move on to the next one.
Murilo:Let's do it.
Bart:Anthropic agreed to a $1,500,000,000 settlement to resolve claims that trained on hundreds of thousands of copyrighted books. The deal packs payouts around $3,000 per book and requires deleting the pirate's dataset spending court approval. For builders, the signal is loud. Pay for quality data or risk eye watering legal bills later.
Murilo:Another big number?
Bart:Big number. And we actually reported on on well, on this lawsuit, I want to say, three ish weeks ago. And I'm very much paraphrasing here, where I understand that the the judge basically deemed, anthropic training on books fair use. So it's okay to train on any book as long as you have acquired this book through legal means. In the years, you bought it.
Murilo:Yeah. So it's kinda like anthropomizing again a bit. Right? Like, it's like you bought the books. You read all of them.
Murilo:You're writing a text that combines all these things that you read. It's fine.
Bart:Well, if you can if you train your model on this, then it's fine. That's what we have
Murilo:to Yeah. But, like, if if I'm a like, if I was a person, it's like, I bought all these books. I read them. I read a text. You put ideas for more.
Bart:So you write a thesis based on this foundation of knowledge that you built through these books.
Murilo:Exactly. Then it's it's that's that's a bit That's right. The vibe. Right?
Bart:The problem is, of course, that the Anthropic did not buy the books. They used ZLip and LipGen, two very notorious book illegal book websites, libraries, very big ones. And that is why they now reach a settlement with a set of authors where they have to pay $3,000 per book. And I assume that these are books from the authors that are in the lawsuit, which is probably a very big group.
Murilo:Yeah. I wonder how they know about all these books. Is it because they just took every book from LipGen or whatever?
Bart:Well, that's what they did. That's what the lawsuit says.
Murilo:So just listed every book that is available there, and they said 3,000 per book?
Bart:I don't know how they came to this number. So it's a good point. No. I I I would assume I don't know. But I I would assume that it's all the authors that started the lawsuit.
Bart:I think I assume it's a group of authors or a group or a group of publishing houses or something like that. I assume it's divided over those.
Murilo:Yeah. See. I see.
Bart:But I'm not too sure, to be honest.
Murilo:And actually, they settled. Right? So it wasn't it wasn't in court. Right? But they settled for 1,500,000,000.0, and I think what they said what I I saw somewhere that they they were afraid like, wanted to not afraid of, but like that.
Murilo:They didn't wanna go to court, and maybe we'll keep hiring, whatever, like the the same story. I also saw that the Entropic had funding round before. The fast growing startup announced earlier this week they had just raised an additional 13,000,000,000 in new venture capital funding to deal with development 183,000,000,000, So I'm not sure if also them raising this money was also to to settle on this. Maybe. Yeah.
Murilo:Maybe. Right. True. And what are the consequences for this? Because I was seeing here.
Murilo:So, yeah, this is not a legal precedent because they settled. Right? But this will also I mean, this also sets a precedent. Right? Maybe not a legal one, but I think if something like this happens in the future, people definitely look at this this settlement and say, this is how much money I'm gonna be asking for.
Murilo:There's definitely consequences.
Bart:Right? Exemplary. And I think I'll I'll let you introduce. And that next topic is I don't think the the timing is a coincidence.
Murilo:So maybe let's just get to it. We have two authors sued Apple, alleging the company trained its AI on pirated books to build its open ELM models. The company complaint the complaint says Apple has not attempted to pay these authors for their contributions to this potentially lucrative venture. For the broader AI industry, it's another reminder. Data provenance and licensing are becoming legal must haves.
Murilo:So, again, once settlement comes, another lawsuit gets filed right now towards Apple.
Bart:Yeah. And this basically comes a few weeks after the the the ruling from the Entropic case that you are not allowed to use, basically, pirated books. You need to buy the books before you can train on them. And what do we have a few weeks later? Yeah.
Bart:A lawsuit for Apple gets sued over a pirated books that they apparently used to train one of their models, OpenLM model. Yeah. It's interesting to see. I don't it's it's it's the first one we see after the the which the sort of precedent that is being set by you're not allowed it's fair use, but you need to buy them. I think we will see much more of these lawsuits coming in.
Murilo:Yeah. I think I think so too. And I think but I I'm also wondering if if you're one of the big players, maybe you can take the hit. But if you're not one of the big players.
Bart:That's a very good point.
Murilo:Right? And I think
Bart:It's a very good point, mate, that you're making. We were discussing Mistall earlier. Yeah. Mistall is is valued at what was it? 11.7 post money?
Murilo:Think so.
Bart:I mean, it's a lot of money. Yeah. I mean, you can't really settle at one point four five and one point five.
Murilo:So yeah. There I mean, I don't know if there was a strategic thing, yeah, but I think there is also implications there for yeah. Like, maybe you're reading out.
Bart:To me, this is a bit like I think there is a very big chance that all the big players use this.
Murilo:I think so too.
Bart:It's super accessible. Like, you can just store on the whole library. Like, it's a I doubt that anyone did not do this.
Murilo:Yeah. I mean, if you look at the things
Bart:a year ago, well, it's was definitely free for all.
Murilo:Exactly. I mean, if you see the the things I mean, OpenAI was was a big case for a lot of stuff, even the image generation. Right? It really doesn't seem like they they they cared at all. They just, like, took everything they could get and just powered the new models.
Murilo:Exactly. And maybe now also the other thing is their models are already trained on this, so they also use the even if they they don't have the data, they don't use the data anymore. They still can use their models to train other models or fine tune their models to other things. Right? So in a lot of ways like the the abuse quote unquote that they they they done can still be reused for other downstream models.
Murilo:Right? Like, benefit in other ways. It's not like if Anthropic trains a new model, they're not gonna start start from scratch.
Bart:Probably not. Right.
Murilo:So yeah. Apple now sued.
Bart:And then also, like, it's probably because now the settlement is 1,500,000,000.0 for Antrovic. Right? It's $3,000 per book. I mean, the books don't cost that much. Right?
Bart:Yeah. And you have to buy them once. That's what we get from the from the ruling it to boy at this point. Right?
Murilo:Yeah. Indeed. Indeed. But that's why I'm also wondering, like, like, 3,000, it seems like a lot, for for
Bart:each That's also why I think it's a it's a small set of publisher or authors that are currently in lawsuit.
Murilo:True. But But that's a
Bart:The actual the LibGen library is huge, and it's it's it's encompasses a lot of different types of texts. It's novels. It's scientific texts. It's tons of different languages. It's I would assume that it's less per book if you would divide over everything.
Murilo:Yeah. Yeah. No. I I would also imagine it's that's also been like how many yeah. So I also was a bit surprised.
Murilo:So what's next?
Bart:Microsoft. Because, Microsoft warned, Azure customers about high latency after multiple subsea cables were cut under Red Sea. Quote, under sea fiber cuts can take time to repair, the company noted, as engineers reroute the traffic to alternative parts. If your apps cross Europe, Asia, or The Gulf, build in Slack and expect occasional real world infrastructure hiccups. Yeah.
Bart:It's a worrisome worrisome news. Right? Geopolitical tensions that we see. I think also the the ability that that that different groups have to do this kind of damage, which has probably has also evolved over time. I think this would have been much harder to do fifteen years ago, or there would be less parties that would be able to do this to basically damage these cables at the bottom of the sea.
Bart:But you see that these these geopolitical tensions, they start having an effect on the World Wide Web. And I think it makes me worry a bit, like, how big can we expect the the Internet to be. Like, if this continues and if this grows, like but we only see we have access to a part of the Internet. Right? And also Yeah.
Bart:That's maybe a far fetched concern, but also, like, what does it mean for like, if you really need to have something that is fail proof. Like, what you typically do is, like, if if you're if you're I don't know. If you're a bank, you have maybe if you're a local data center, you have a multi cloud setup, You are you have different geographical zones, but maybe now you also need to have different network routes to connect to all those those things. Like, it's a it's a big extra worry if these things, like, grow in in in occurrence. Right?
Murilo:Yeah. Indeed. It's a bit yeah. Indeed. Like I said, I think it's a bit worrisome.
Murilo:Right?
Bart:It's a bit worrisome to like, because it's to me, like, this is like the Internet is a bit of a should I put this may maybe a bit of a basic necessity
Murilo:Yeah. For
Bart:for the the human race as a whole
Murilo:at this point? It's like a commodity. Right?
Bart:It's become a commodity. Yeah. Yeah. But this put it a bit at a at a this puts it a bit at attention. Right?
Bart:Like, we hope that it will remain a commodity.
Murilo:Yeah. Indeed. I think I think there's a lot of the time is also there's a lot of tensions in the world. Right? And I think when you see these things as well, it becomes very real, I guess.
Murilo:Like, it can becomes very close to home. Right? I think sometimes it feels like it's very far.
Bart:And and and I think you had this to some extent as well. Like, you have regimes that really close off their parts their country's Internet. Right? Like, if you look at North Korea, but it's it's a bit more more isolated. Right?
Bart:Like, this is really disruptive for global communications across the World Wide Web.
Murilo:Yeah. Yeah. Indeed.
Bart:So, yeah, let's let's hope we don't see this too often.
Murilo:Let's hope it's the first and last time that we have some articles like this. And next up, what do we have? We have OpenAI. OpenAI says hallucinations persist because conventional benchmarks reward confident guesses instead of admitting uncertainty. They argue evaluations should penalize wrong but confident answers and sometimes credit I don't know responses.
Murilo:For users that means more calibrated models that hedge less often and bluff a lot less when stakes are high. Did you get a chance to look at this article, Bart?
Bart:I browsed through it. Yeah.
Murilo:I browsed through it as well. What caught my attention the most was the conclusions at the very bottom. So maybe hallucination in case you're for the people that are not familiar, the way I would describe it is when models they say something that is completely nonsense, non factual, is just wrong, but they sound very convincing. Right? They sound very confident.
Bart:And what what they're posing here is that they're they're saying that this is because the benchmarks that they they train and test against, basically, is that they penalize saying, I'm not sure.
Murilo:Yeah. Yeah. Also
Bart:not sure is It's a bad answer. Sounding very confident.
Murilo:Exactly. I think also in the way that these models are trained, right, with the human reinforcement, right, That even in that period as well, people reward very confident. Right? And I think I'm
Bart:Maybe there's also this bias because it's
Murilo:I think so.
Bart:Human reinforcement.
Murilo:I would guess. Yes. I think it's the same. I mean, again, I'm working consultancy now. Right?
Murilo:And I do also know the the power of confidence.
Bart:You need to sound confident.
Murilo:You need to sound confident. Even, I mean, even when you're kinda not. Right? Or, like, when you I I don't know. It's like, it's I know it's it's not perceived well to not have an opinion as well.
Bart:Mhmm. Yeah. Exactly.
Murilo:So it's like having an opinion, being confident, being able to black it up. I think it it it goes a long way, and I so I think there's also something human there. So I also imagine that in the human reinforcement learning that was also played a role. So I also think that's why models are they were tuned to be very confident. I think so models, they are predicting the next token.
Murilo:Right? The next word. And so I think, yeah, that's the main the main takeaway. Right? They say hallucinations are the claim is that hallucinations are inevitable, and actually, they they say they disagree.
Murilo:They say, actually, they're not because language models can abstain when uncertain, which I mean, maybe there will still be some hallucination. Right?
Bart:Yeah. Because that's that's a strong that's a strong finding. Right? Like, they're saying that hallucinations are not inevitable because the because the model can abstain when uncertain, but it also assumes that there is a good this good estimation, like, when is this when is this model uncertain?
Murilo:Exactly. And I think this also links a bit to another another finding that I thought it was pretty interesting that avoiding hallucinations require a degree of intelligence, which is exclusively achievable with larger models. Next, you say, not necessarily. It can be easier for smaller models to not hallucinate, to basically abstain. Right?
Murilo:So the example they give is, like, if you have a Maori question, a small model that doesn't have anything on Maori, they'll just say, I don't know. But a model that has a bit of information, they'll they'll just make it up because they know some of it. You know? So I also thought it was quite interesting. And, yeah, I think that, like, you mentioned, like, other things about the the measuring hallucinations.
Murilo:So I thought it was an interesting article. I also think that today, that's the main for people using AI, I think that's the main caveat. Right? You always need to be a bit like, maybe they don't know. They're saying a lot of stuff.
Murilo:They're always gonna give you something. If you say do this, they're always gonna do. Even if they have no context on how to do, they'll just get something, which sometimes is good. Like, if you have writer's block or something, you just wanna get something on the paper, maybe that's good. But for a lot of the things, it's actually pretty bad.
Murilo:Right? Like, maybe if you're if you're asking the the model to come up with a solution, a technical solution of something that makes no sense and it's impossible because it it it doesn't it refrains itself from saying I don't know. It will come up with an answer, but the answer will be nonsense. Right?
Bart:And it will be very sycophant.
Murilo:Exactly. Exactly.
Bart:Great question. You're the man.
Murilo:Yes. Do this. Yeah. Yeah. Exactly.
Murilo:Exactly. So I think that's one of the main I mean, for people that actually use this quite a bit. Right? That's like you for learning or for any other case. I think that's the main thing.
Murilo:Like, you need to always be like like you mentioned, you're driving, and it says, ah, this and this and this. Like, okay. Give me sources. Like, you have to have that muscle. Right?
Murilo:And I think if you can if they could improve this in future models, I think it would be a huge, huge, huge step forward. Right, to just ask follow-up questions instead of just giving answers.
Bart:They improved a lot with ChatTripety five. This is Okay. Personal my personal experience with ChatTripety five. Yeah. And but, also, you hear them a lot on community.
Bart:GPT five, made a big step when it comes to, avoiding hallucinations. Still hallucinates sometimes. You also have these two different you have this, instant response and the thinking response, and the thinking response is also very good at this. It tries to
Murilo:Okay.
Bart:Typically, like, find online resources to back up a statement.
Murilo:Okay. Cool. And then have you ever do you encounter more in CHIPTY five? The the the moment where it says, I don't know or not as much?
Bart:No. I can't really say that now. No. I can't really say that I've noted.
Murilo:Yeah. But I do think improving this would would drastically improve the experience people have with the large language models today.
Bart:Definitely. Definitely. Yeah. Yeah. A very important problem to solve.
Murilo:Fully agree. Alrighty. I think that's it for our tip for today. That's it for today. Indeed.
Murilo:Oh, well, that's it for today with an asterisk. We have some tidbits.
Bart:Tidbits. Let me just quickly go over them. I think one interesting one is that AMD claimed that that ARM, the e r architecture, which is, for example, used by by Apple's CPUs, it doesn't offer any efficiency advantage over Ticks are the the typical CPUs that we architecture on CPUs that we we have used for a very long time on almost all devices until, Apple switched to ARM. The moment that Apple switched to the m one on the laptops, like, people suddenly had a huge amount of battery. Right?
Bart:Battery life. I could switch from two hours to eight hours, something like that. That's a bit of the feeling I had when I look back. And I think everybody was talking back then, and to me, that was still my belief as well. Like, the main thing that made this is that ARM, ARM, is just way more power efficient.
Bart:And, apparently, it's just a combination of things where Apple optimized a lot of things, on their OS to be power efficient. Also, the way that they integrated things on their, on their circuits made that it's very power efficient. And and the is saying and, apparently, it's also something that is, like, quite well known under under under experts that x 86 is as power efficient if you want it to be. And I think it's interesting doing I think because I honestly didn't know. Thought I honestly did but then the big change in battery life when it comes to, especially Apple laptops, was the switch to ARAM.
Murilo:Which makes what do mean? Which would make sense. Right? Like, they switch to m one and then the battery goes up. I think it's a but it's a good example of correlation, not causation.
Murilo:Right? And
Bart:Yes, sir. Yeah.
Murilo:Is there, like was it more a marketing thing then that they switched chips, or are there other benefits if it's not power?
Bart:Good question. Definitely not the the person to ask in-depth questions on this.
Murilo:We will look for if anyone listening that knows the answer, hit us up, and let's have a chat.
Bart:The other one is something from the UC Santa Cruz news, told an interesting article that I spotted. I think on Hacker News, it's that Wi Fi signals can measure heart rates, and you don't need any wearables. And this is just cool. Right? Like, it's probably not it's probably far away from being practical.
Bart:But, apparently, like, the you can have your Wi Fi sensor. Like, your Wi Fi sensors have very frequent signals, and it's what they do is that they record what comes back from these signals, basically, quote unquote, echo. And from these patterns, they can and when they remove all the noise, they have a very accurate estimation reading on your actual heart rate. But, of course, there are limitations. Like, you need to sit quite still.
Bart:You need to be the only person in the room. But Mhmm. I mean, this is just cool. Right?
Murilo:Yeah. It's cool. It's pretty I actually saw a while ago that you could use what was it? You could use Wi Fi signals to kinda see through walls, and, like, you could kinda I don't know. The the use case that I remember is, like, if there's an elderly person in the house and they fall, you could actually see that with Wi Fi.
Murilo:And I was like, that's pretty cool.
Bart:Became a bit X-ray vision. Right? Like,
Murilo:can see through Yeah. Yeah. I was like, that that I mean worrisome as well. But A bit worrisome, but I was like, let's go still. Yeah.
Murilo:That was actually some years ago, I feel.
Bart:So That's true. Yeah.
Murilo:But, again, I do feel like for for older people living by themselves, I think it's a it's definitely a definitely a use case. So one of my neighbors is a is older older woman as well.
Bart:And you now have your Wi Fi sensor in the thick at the wall? A t shirt wall?
Murilo:Just like yeah. We we all know them are monitoring to make sure Exactly. Exactly. Everything. Yeah.
Murilo:But, no, we all know them are Wi Fi. It's not the best part. So, anyways, what is the last one? What's this who's the author of this?
Bart:What's this? Small tidbits. Article that I posted myself on my blog.
Murilo:Wow. Bart's that's you. Bart's space.
Bart:A bit of a thought experiment.
Murilo:Nice nice spaceship here. That's you.
Bart:Nice spaceship. Yeah. So that flows over across the screen in my blog. But we can check it out from, last week, we were discussing the bare block switch, from their MIT license to, Elastic license to, from Yes. Open source to source available.
Bart:Yes. And I was thinking, like and and, again, it's a bit of a thought experiment. Like, don't you need a different way to look at this? CloudFresh looking at at, like, blocking crawlers. Like, could, like, a large platform to, get up, not do this?
Bart:Like, even, like, frustrations, like, my code is being used to train a model that can just make a copycat of my code.
Murilo:Right?
Bart:And I think at least you should be able to say, want to block crawlers from using my code at GitHub. Yeah. So I and probably, like, the resulting model, if it does dev access to your code, can still create copycats, but at least it was not trained on my own code. Right? Like, it's a bit of a principle thing.
Bart:Maybe it doesn't really make change in the bucket. But if you if I have a very popular repo or I'm redoing some novel stuff, maybe I say it's okay to crawl my my my repo, but just pay me. Right? Pay me or or get the fork out.
Murilo:Right? So it's the MIT PMO GTFO license.
Bart:Yeah. MIT pay me or get the fork out. And I think HitApp should adopt this. I think that the very difficult thing, of course, is, like, how do you block roles. Right?
Bart:And how do you block roles from correct correct derivatives from your code? Like, it's a it's a very, it's a very complex thing to achieve, but I think, like, with with players like Cloudflare really going the next mile on this, like, this is a realistic future where you should have the option to block crawlers or get paid for it. True.
Murilo:True.
Bart:Also, vote that I just just saw you clicking.
Murilo:Yeah. Yeah. Yeah. No. It's a it's a nice well, I just got the TLDR, let's say, but it's an it's an interesting concept.
Murilo:It's a nice idea. This doesn't exist then. It's not something you saw somewhere. You were just saying, like, why don't we have this for code?
Bart:Yeah. Why don't we have this?
Murilo:Yeah. It's true.
Bart:It's actually the the bare block discussion that made me think about this.
Murilo:Oh, and and I agree that it's it's it's fundamentally the same as, like, books or it's similar. It's related to people using open source projects to make money. But why do you think I don't know if anyone thought about it before. Right? If they have, we haven't heard about it.
Murilo:Why do you think that is? What do you think is is there something different about code that people just expect it to be taken, pulled, and copy pasted? Or do you
Bart:I think With code, if you make the comparison with a book, I think, like, if I have a I've built something. I've built an open source product, and I'm hosting my own service on that. I think BearBlog is a good example. They they built a blogging platform, and they're hosting it, they get some revenue from that. They're they're very passionate, and they build something.
Bart:The problem with code is that it's with code and especially with Jenny I now, it's very easy to create a copycat of that even if it's just source available because then you say to the GenAI, go through this repo, make sure that it's different enough so that people don't see if there's a direct link.
Murilo:Like, one can
Bart:be the different can be the same. So it's very easy to make a copycat and guess more or less like like a a similar type of SaaS maybe with your own with your own a bit like UI UX is a bit different. Maybe if I if your own niche that you're focusing on, but it's very easy to make a clone and profit off of that. And I think with a book, it's way harder because it's like it needs to be like, even though you maybe train on the content of an existing book, but, like, to be successful on in a book, like, you need to have, like, a substantially substantially different different book. Book.
Bart:Like, Like, I'm I'm not not gonna gonna just make a clone of the Lord of the Rings. Right? Yeah. I mean, it's probably not gonna be very popular. I'm probably not gonna profit off of that.
Bart:So I think there is this difference in, like, don't just take my work that I worked so hard on and and take away my profits. That's a bit of the
Murilo:Yeah. True. Still one bit.
Bart:The combination of those things.
Murilo:I feel like when you when you pose the question and you suggest the the new license, let's say, I'm like, yeah. Of course, we should have this. I'm wondering why I had thought about it before or why I haven't heard about it.
Bart:The difficult thing of this is, like, having a license there is easy. Right? But enforcing
Murilo:this super hard. Yeah. Yeah. Yeah. I mean
Bart:Yeah. I think the only the few players like Cloudflare can maybe achieve this.
Murilo:Cloudflare. But for Codi for Codi, you're thinking for GitHub. Right?
Bart:Well, yeah. But I mean, like, understanding, like, this Internet activity is actually a crawling bot from one of the big AI operators. I think that just understanding that this is super hard because, like, there's a lot of abuse. Right? Like, it's
Murilo:Yeah. Yeah. Yeah. I see what you're saying. Even if GitHub wanted to do it, I mean, you can put the thing there and maybe GitHub will respect it to not train Copilot there, but how wouldn't Tropic or or anyone, like, they can just afford to go and
Bart:just to hope that they that they abide by these rules. Right? And maybe and then GitHub is a big enough player again to have some bargaining space with these operators. Right?
Murilo:Could they also enforce this on OpenAI or Intropic? Or
Bart:Well, GitHub is Microsoft. Microsoft has a very big stake in OpenAI. So maybe they can do it on OpenAI.
Murilo:Maybe they can. That's true. That's true. Alright. Cool.
Murilo:Actually, this is this is a very good idea, Bart. I think I think I think I hope this gets the right years. I think this is a it would be nice
Bart:to see
Murilo:it would be nice to see this going forward. I think this is something that we're missing indeed. Cool. And this is bart.space for people that want to check it out. Interesting read.
Murilo:Thanks for sharing, Bart.
Bart:You're welcome. That was that was it for today already.
Murilo:That was it. I don't know. I feel like I feel like we had a lot of topics, but we got through them. Got through them. Cool.
Murilo:And anything anything else you wanna say? Any parting words? Any any inspiring thoughts?
Bart:Go hard or go home.
Murilo:Wow. Okay. No context. Just that.
Bart:The best without context.
Murilo:Okay. Okay. Cool. No questions. But thank you.
Murilo:Thank you, Milo. Thanks everyone for listening. As always, if you enjoy, feel free to tell a friend, leave a review, hit us up if you if you have any feedback, anything.
Bart:It's just to make it to specify a little bit, leave a five star review.
Murilo:Leave a five five star. And, yeah, if you would like to join us as well, hit us up as well. Have any feedback, anything. Happy to have a chat. Thanks, Bart.
Murilo:See you next week.
Bart:See you next week.
Murilo:Ciao. Ciao.
Creators and Guests
