ChatGPT "After Hours"

JR Reynolds: This is Yelling at Robots European Edition. We'll get into it, but first, a Yelling at Robots episode five. ⁓ Here we go. Action. Get ready. It's time for world famous Yelling at Robots. Starring tech legend, Joshua Raphael Reynolds and Ian Thelonious Bosberry. The show starts now. Welcome to Europe, my friend. They let you in. They did. Yeah. Tell us where you are. Tell us what's going on. So I am currently at a Nobu in Warsaw, Poland. I'm here for work. So I've been here all week. And you're just now adapting to, I mean, how's your jet lag? You're just now acclimating? Yeah, so not to be too graphic, but yeah, I think that my old body is finally. I pooped in the morning today. Were you in the bathroom at the time? is that? Cool. So this today's episode. Yes. First off, we are on a tight deadline. I forced Ian to get up ⁓ and interrupt his pooping time to join me in the morning because he's going to be on a flight in the afternoon and we normally record in the afternoon. ⁓ And we also are trying a new experiment today, which is to prepare. ⁓ I've got my own notes, believe it or not. You your own notes? Wow. Wait, notes that I can't see? Yeah, yeah, like notes, my thoughts about our topics. Oh, interesting. In my org mode, Emacs. Oh, God. Yes, I still use Emacs and I still, well, I only use Emacs for org mode and taking notes. Leveraging like one-tenth of the power of Emacs. Yeah, but I mean I'm not doing it. Well, I promise This is just make an agent do it for you Emacs agent. ⁓ man, that thing will be quite mad. It will be quite insane It'll be like open-claw it'll be it'll basically be like wait I can fix this delete You know, I've also been on vacation for the last three weeks are kind of not not producing While I was on vacation, I cut our episode where we actually looked at some real code that was fun and that was a very beefy cut. But one of the things that I, this is a very selfish personal thing that I noticed was when you were doing a podcast, you really notice how many times you say like you just pause. And Ian, you are naturally very good at not doing that actually I've noticed. You curse like a fucking sailor. ⁓ yeah. Yeah, yeah, yeah, you've probably noticed for those of us listening, or for those of you listening, pardon me, that Josh only catches his own curses with the bleeps. And a few of mine fall through the cracks. It's hard. mean, I have It's a really hard job when you got old fuzz over here, just fuck this, fuck that, fuck it, asshole, fuck, fuck, fuck, fuck, I have to do a whole edit pass that's just fuzz cursing. it's like, the episode's 45 minutes long, it's allowed to work. Two hours to edit the episode, four hours to catch all the f***s. Four hours for profanity management. Can we get an agent? I'm also working on a profanity management agent. Yes. We've been off for three weeks and three weeks in AI news is a year in the real world. I was thinking... I was ⁓ today, March 27th. So the first one we're going to start with is an article or post, what have you, from a gentleman named Adam Bezemek. I hope I'm pronouncing that correctly. I just call him Adam B. My B-boy. The title of the post is From Builder to Orchestrator, the Identity Crisis No One's Ready For. So in it he ⁓ kind of talks about what he's seeing in the field ⁓ in terms of ⁓ You know, they kind of off the top. The big the big highlight poll quote would be that Boris Boris Cherny, the head of Claude code, at Enthropic recently told Fortune that he hasn't written a line of code in over two months. He's shipping 22 to 27 pull requests per day. All a dick. Yeah. I actually quick note, I had a conversation with someone while I was here who was doing the same thing. And one of the things they said was my team can't keep up with me. And I was like, yeah, no, of course not. Nobody can keep up with that. And so he talks about things in this in this post about like this, this feeling for a lot of engineers, you know, especially ⁓ Josh and myself, which is actually probably one of the reasons Josh said, hey, maybe maybe we should do a podcast about this is ⁓ this can cause cause an identity crisis ⁓ that nobody that nobody's talking about. Right. We became engineers because we like building things with our hands. He also mentions this recent study that found AI assisted coders, like, sorry, what's the word? No, sorry. ⁓ This is a kind of different topic in the, the, in the post that talks about, what is it? Like a skill erosion, right? And this is a thing that we've always talked about, which is just the more that we use AI, the dumber I get and the more I feel like I can't ⁓ Which I think I've found a balance with that now. That was more early on when I had started using it and getting it to solve problems for me. I would basically tell it the problem, shut my brain off, copy paste the answer, that sort of Wait until it came back. Yeah, but that was before we had agents and things that could actually edit our files. That was just when you were talking to ChatGBT. Oh, sure. And that's real. the study that he references, which is weird because the study, I read the abstract and I actually like, glanced over it, but the study is about something totally different, but I guess in this study it does talk about AI assisted coders scored lower on mastery quizzes compared to those who continued to code by hand. And that's like two letter grades in school, right? ⁓ Which is not insignificant. Were ⁓ the AI coders allowed to code with AI? were they? the AI coders. No, no, I don't think. Yeah, that's funny. Well, yeah, if I'm an AI coder, you take away my AI. It's going to suck ass. Right. I don't know. I didn't I didn't fully read that thing. I was also confused. I'm like I started I read the abstract and I was like in nowhere in the abstract as it mentioned this study this this mastery quiz thing. And I'm like, wait, OK, OK. Anyways, that was a bit confusing. OK, but I'm sure it's true because there are. Yeah, I think really famous. a really famous study from last year, I think at some point, that did actually study ⁓ seasoned open source contributors and had them estimate how long it's gonna take me with AI, how long it's gonna take me without AI, and then did a control group, split 50 % of the issues, and lo and behold, takes longer. With AI, takes longer. Anyways, there's things I agree with in here. I because you're not using it enough. You're not going hard. Yeah, yeah, yeah, mean this is counter argument, right? Which is like we're trying to fit a square peg in a round hole dummy. You need to go all in there's things I agree with here and there's things I don't and I I think that the like the things I agree with like he does go on to talk about you know things you can do to help you deal with this feeling of identity crisis and these feelings and stuff like that and he gives you a bit of a framework and stuff and things like, you like for the skill erosion problem, to deliberately try to preserve your craft, right? So for instance, I know you asked me in text, like, how did you use AI this week? And I actually didn't get to, because it was a lot of meetings and collaboration and stuff, because I'm here with the team. But one of the things I was doing was one of our repos that manages a lot of our pipelines is actually pretty heavy on the FPTS and functional programming and I hadn't been in there in a while so I couldn't remember like what's a bracket again and what's the difference between either in a task either and so I was using AI to reload a lot of functional programming concepts into my brain or FPTS concepts into my brain. guess both. And just sorry just to refresh my memory FPTS is a library suited to enable functional programming in TypeScript or? Correct. It's opinionated and it has its own kind of medicine. It's very opinionated. It's very, we, we would joke a lot on the, when I was previously at reactor, ⁓ we would joke that the docs are very much. Learn category theory or fuck off. ⁓ okay. Right. Right. Right. Sure. Sure. Makes you feel super smart once you wrap your head around it, but yeah, there's a bit of a bit of a blocker. Anyways, back to the, back to this idea of. things I agree with and things I don't so something else he said was that one of the one of the other I don't know pointers or maybe the last pointer was Except that this might not be your Except that your time is coming to a close and the Sun is setting on the land in which you can live. Yeah I would I would say about that is I won I believe That's a bit of a false premise because that's kind of accepting that everything you've said prior in the article is fact. True, right. It ain't, but we're not there yet. Like, yeah, you're seeing all these things and people moving quickly, but like a lot of that is driven by a lot of bad decisioning and stuff like that, which I think probably leads to our next topic. I was thinking about it too, because it's so easy. It is so easy to get. like FOMO'd to death when you when you look at what's people are writing and people are saying. As I've mentioned before, Twitter is a... I open up Twitter and I'm like, I'm fucked up. ⁓ But all these, like everyone's got an angle and got something to sell too, right? As well as I'm thinking about kind of ⁓ Andre Carpathi, right? I think about Andre Carpathi a lot. That dude went from being like, we're not there, LLMs or it's all slop to like a month later being like agents of the future, we're all fucked. It's done, we're done. And I was like, what? I was thinking about kind of, some of the, and he's obviously hugely influential and everybody follows his Twitters and is just like word of God kind of thing. And I was thinking about it and it's like, dude is also just a dude. So like, he's right about some and he's wrong about some. He's just a dude. He obviously is in a much different position in terms of seeing things and having knowledge, but also some of the anecdotal things he's referenced that like software development is dead. Look, the, Old ways of working are indeed dead. We need to adapt and change and learn, but it's not, as a craft, as a skill, we don't know what that means. We have no idea. Well, we might have some idea what that means. But his anecdotal references were like, he vibe coded an app to manage his security camera at home, which in the past he would have peddled on and taken him three weeks, and he did it in an hour with prompting and did no introspection, inspection of the code, and it works and it serves his needs. For me, that correlates almost not at all to how the world works. That's great. something's wrong with that app, how many people does it affect? It affects him. Exactly. It affects him, and it's not interconnected. It's not interdependent. It's not part of a larger system of commerce and interconnected systems and systems and systems of the world. so that's where I think my position is is still wait and see. Yes. Moving on, I would. Yeah. Yeah. The next article was my favorite thing that I read in the last little while, but only because this article by Gurgley. ⁓ this guy's he's been writing pragmatic. He's been writing a lot. So he runs he runs his own subject called the Pragmatic Engineer. He's wonderful. He's got a he's also got a accompanying podcast, which excellent, excellent. ⁓ episode that I listened to recently which he interviewed Mitchell Hashimoto the the founder and The founder of hashy Corp and you know now the the leader of the ghosty team ⁓ Both both technologies which I've used extensively ⁓ Was is is is getting early the guy who was a was he an engineering manager at uber or I can't remember Yes, he's been he's definitely he's been at multiple Silicon Valley like he's he's he's he's been around the block and he knows the shit Yeah. Cool dude. Cool dude. Great connections too. get a lot of, and a good, Enjoy his work. And a great podcast. And Gargeli, if you are interested in appearing on Yelling at Robots, we will drop contact information. I'll reach out to you on LinkedIn for our next, maybe our first guest, but it'll probably be like, she's like, can't say my name, Gurgling Motherfetcher. f*** you. It just seemed fun. I'm sorry. I've never heard your name spoken before. That's OK. So anyways, he this article called our AI agents actually slowing us down. The answer is yes. And it was a very short article. Yes. It was one of these seven parts. No, it was actually very well researched and thought out. And he's referencing both ⁓ data and opinions from other leaders in the field ⁓ who are like still in the trenches. I really, and so the first few things he points out are like the, like right off the top, the first example is like the degraded flagship website of Anthropic itself. And the example he shows is, if you dig into the example and you have built any front end thing in any capacity, you'd be like, how did no one catch this bug? This is insane. Yeah. Yeah. ⁓ So that, yeah, there's that, there's the increase in SEVs or severity incidents at Amazon that they are like, well, juniors always need a senior to sign off on their work now, which, yeah, that should solve it. He goes on to talk about... let's pause there just in case people, I mean, there is so much news that comes out. So this Amazon adaptation is a result of... a couple very, very serious outages at Amazon. They call them SEVs in their parlance. ⁓ One that was caused by an AI agent that was doing automatic, I think it was automatic infrastructure configuration and whatnot. That was some time back. And then a recent one that was a junior engineer who took mistaken advice from an agent that was fed data off of an out of date, like, you know, confluence space, and then was just like, well, agent said, bloop, and caused 13 hours of downtime. for a credit, I don't remember what service it was. It was ironically for their version of cursor or their AI coding ⁓ thing. ⁓ If I'm not mistaken. You might have to check that. As a result, so Amazon then instituted a policy that no junior devs are allowed to ship code without senior checking off on it. Which also, the other relevant factor, and I don't think that it's in our ⁓ in our agenda is that Amazon in an absolutely insane, like have we learned nothing in the last 40 years, it uses token count, token usage as a performance metric for engineers. Were you aware of this? ⁓ was Amazon doing that? Because I thought it was just Meta. Well, I guess Meta, no, Meta and Uber are doing it for sure. So that's point three that he talks about in this. Amazon's probably also doing it because if one does it, they all start fucking doing it. think I believe so. mean, I guess I could be mistaken. check. will absolutely not check that after the episode or put that in the show notes because I'm too busy. But just take it for true. These are like managing to the numbers and optimize. They're definitely doing it. Metta and Uber. There are teams doing it. Yes, there are engineering managers at Metta and Uber. tracking token usage and that is a metric that will be on your performance review. How many tokens are you tokening? Are you maxing, what is it called? Token maxing. Token maxing. I want a t-shirt, token max, token maxing. I'm sure nobody could game that. Well, it's just ⁓ kind of like learn nothing from bad management practices in the last. 50 years, 60. We've known these bad management practices just like either basic like economics and like incentives. Yeah. So so carry a book. Read a book. Read any book ever written. OK, so back to back to get a galley. Yeah. So he talks about that. That was like the third section. And I would encourage anyone who hasn't signed up for his. for the pragmatic engineer to check it out. great. But one of my favorite section was because it's a tool I use, is OpenCode. OpenCode is a open source terminal based orchestration way to speak to your whatever your LLM of choice and use it in your, like so I personally, because I do terminal based work, I prefer to stay in that. That's part of my workflow. didn't want to like. start leaving to have to figure out how to use VS code or cursor or whatever. So OpenCode gave me way to do that and it's really configurable, well designed, it's great. It's kind of like Claude CLI but like more flexible. Although they did It's Claude for dorks. Yeah, yeah exactly. For open source Claude alternative, agentic harness, coding harness, managing all of the skills, agents, and shares. And we'll get into that actually when I talk about what I did in AI this week. Sure. But yes. Indeed, OpenCode. Check it out. So OpenCode's great. They did get in a fight recently with Claude, and Claude has said you are no longer allowed to connect to our API. So, god damn it. Anyways, but OpenCode, the actual team, he references something that DAX, the team lead, whatever you want to call it, like the guy who started the project, warned about, which is just that the, and this is really interesting, that the... Not a lot of people think about this. It's so easy to make a feature these days. Or a ship code that no one stops to say, should we? Should we do this? Yeah. And think that... And then that also, that dopamine hit and that speed at which you can pump this stuff out, it discourages revisiting for refactoring and taking a look at... well, what we've built over the last month and is any of this still being used? And you can just slowly, bury yourself under code and probably technical debt. So that's another interesting thing that he talks about. The CTO of Sentry has actual metrics. Sentry is a, for those of you who don't know, a, like a New Relic or a lot of these tools, it's a instrumentation tool that developers and teams use to measure incidents that happen in their applications and then debug them and figure out what's wrong and where in the code something exploded yada yada yada. So it's a very good tool. We've used it at Paramount. The CTO of Sentry has actual data that says that it is producing bloated hard to maintain code and it has slowed down long-term velocity. Right. Cool. yeah, I didn't call that one. Feels good though. Feels good, man. Feels good. Feels good. good. Tell the bot to do it, then get it done, and then ship it. Feels good. Done. I'm already done. Well, the thing that Yeah, but the fucked up thing is that we all know, like, it does, when teams start to adopt this stuff and move quickly, like, it does, you get short-term gains. 100%. Right. Oh, velocity's increased over the last month. Surprise. Right. Well, when we're doing 22 to 27 PRs a day. Yeah. Well, I have, I don't know if you remember, Fah, a couple episodes back, I christened Shower Thoughts. Shower Thoughts. I have a shower thought, which is, it's generative AI. So, indeed, it speeds up the generation. We're speed up output. It doesn't mean we're speeding up, and maybe this is kind of coming back to what Gergely is talking about. It's not speeding up progress, necessarily. It's speeding up output and generating more and more output, which I think you can't make a case that that actually will slow down progress when pieces of output need to interact with each other or do things or like blah, blah, blah. The product side of things is also a really interesting one. And I experienced this because I am a Claude user. I'm a pretty committed to the Claude ecosystem, Claude code. But although we'll come back to that. ⁓ Their product sucks from a product standpoint in terms of, well, no, just like in terms of it is, you know, ⁓ well, you can do cloud chat. can do cloud coworker. You can do cloud code. Now there's three tabs all in the desktop application. When should you use which, what extensions and plugins should you use? There's cloud skills. There's skills. can install a skill and there's a skill marketplace, but there's also plugins and you can install plugins. There's a pocket marketplace. But then you can install MCPs and there's an MCP marketplace. And it's like, how do I use this tool effectively? Like, what should I do? What's kind of best practices or what's going to help me use this tool? Oh yeah, they're not nudging you in most effective way. No, they're not helping me be effective using the tool. And they're introducing so much change on such a regular basis that it's like, you know, me as an engineer, it's like, I want to have a repeatable environment in which I can create the work processes and practices that work the best for me. But that needs to be repeatable. I need to understand it. And I need to be able like. If my computer crashes, how quickly can I recreate this environment? And the way that the tool is currently built right now, I think it would be really hard. It's much more moving backwards to like you lucked out on a pattern and a path and the right MCPs enabled and the right skills enabled so that it doesn't get confused, so that it's picking the right tools when it's doing the things. But it's completely alchemical. It's not. engineered, it's not actually repeatable. And that for me is also like, it's at least connected to this kind of, ⁓ know, we need a marketplace. Claude, build a marketplace. It's like, ⁓ well, hey, how about plugins? Plugins is something people talk about. then as well as kind of the pace of change is so much that I think it's a disservice to the users because users take time to learn. anyone who's worked on a product. 100 % What's to learn a pattern? And it's like, I don't want this thing to change very much because that makes me adapt and change and change is hard and painful. So... I think your point though, like your point earlier, that it's not even... and this is at multiple levels, like, you know, the wait and see. Like, adopt these things slowly or whatever, but like, I think that... What am I trying to say here? I don't think they know what the patterns should be. Definitely not. Of course they don't. Nobody can know. spaghetti at the wall. Nobody knows. Nobody knows. knows. This is all emergent practice. There's so much pressure to deliver the return on investment that there's not actually time to f*** down or play or experiment or you know it's like if you've got your performance review coming up and you need to be token maxing, it's like, well, I actually just wanted to doodle around a little bit with like, you know, a skill. want to run around with some tokens. I just want to doodle. And the doodling is not going to take make me use very many tokens. ⁓ There's another thing that the article mentions, which is interesting too, is after studying, it says here, 807 open source repositories on GitHub. This study from Carnegie Mellon. talked about that velocity gains and commits and lines of codes are short-lived and tech debt blah blah blah blah blah. But one thing that was really interesting was that AI seems to write more complex code than humans. That was... And compound that over time until you need to go in there and fix something yourself. Yeah. And it's not gonna be, it's not gonna be great. So it's maybe going back to the orchestrator, the idea of being an orchestrator, part of your job as an orchestrator is to make sure that... things are being done in a way that you want them done in the way that you would expect and not just letting the agents run wild for a few months until. Well, there's this, mean, this is kind of the open-claw premise, which is, and this is agent, the whole agent design, give it a goal and give it tools and turn it loose. and it will eventually, if it has the capabilities, will eventually achieve its goal. Or that's the premise, right? Okay, so this was, yeah, yeah, yeah, and this was the argument that the friends of mine made, or one of the arguments he made, and he also talked about, he's built his own little agent framework where he gets these things to fight each, not fight each other, but argue with each other. So one will do something and then there's like a red team agent that comes in and says, this sucks and here's why. And they go back and forth. And they'll actually do this in PRs too. yeah. And that's really interesting. his good, his point was if I can move and I don't know that I necessarily had a, I mean, I think there will be long tail effects to this, but he did make the point that if I if I can, the cost of being wrong has gone down a lot, right? So I can, I can, if I can move quickly and be wrong nine times and then get it right the 10th time, like I wouldn't have been able to do that before this. That would have taken a human a long time to do that. I mean, I probably wouldn't have been wrong nine times. Maybe. Yeah, but what does that mean? I think my question in that scenario is like, what does it mean to be wrong? Is it being wrong while you're deciding what to commit? Like what to actually stamp into a version or is it being wrong three months from now when both you and the agents have lost all context and understanding of the problem at hand or I don't know. It is an interesting like, I just. So I think that's the counter argument, right? Which is... Yeah, indeed, maybe agents generate more complicated code. What's the solution to that? More agents. More agents. going to have five agents that are like critical dickheads who are pushing for simplicity. then we'll have one agent who's like, fucking neck beards. And meanwhile, NVIDIA is just rubbing its hands with glee, right? Because even there was, oh, I think actually, I said GTC, whatever the NVIDIA conference. happened while we were on our break. And I mean, it's really clear. Jensen Huang is essentially just this is the tap. Tokens, intelligence is the tap, right? And it's like, use more intelligence, pay for intelligence. Intelligence is a commodity. And it's like, and they are, they control the pipes. And scam altcoins out there saying the same thing. I texted you about that. I was like, you did. We could have, did you get my reference? What was the text? God, for context for anyone listening, Sam Altman said, he basically talked about the future where intelligence is a commodity and we pay for intelligence. And I was like, Jesus Christ, we could have had the Federation of Planets and instead we're getting Weyland-Yutani. ⁓ yeah, Weyland-Yutani, which is of course the dystopian ⁓ company from Blade Runner, I believe. No, from Alien. From the Alien universe. Oh, even better. Even better. Oh, it might be in both, actually. I guess it's both Ridley Scott. Maybe Whaley and Utonis also I it actually is the same company. There you go. It's such a megacorp that it spans universes. And then, people are just oppressed and put into debt for this one company that everyone owes money to. Yeah, yeah. ⁓ sorry, you want more intelligence. Like you need to earn your intelligence tokens by heading down to the sex farm. I don't know how you earn intelligence tokens. Get down to the docks. Start moving boxes. Those little, those little nimble fingers are still good for a couple things that the bots haven't figured out. Dark times, my friend. dark times. Anyways, I would encourage everyone to read that. Read that article. Good article. Incredibly well done. Great article. Well researched. Confirming most of my biases. Always the best. The best articles. The best articles where you're like, ⁓ my god, I was so right. I was right. Once again. It's hard. It's hard to be Fuzz, Ian Fuzzberry. It is. And be right so much. Okay. All right, buddy. Next topic. Buckle up, folks. After hours This is I was so excited to talk about this one and I did I'm so glad that I haven't I didn't know about this I hadn't been following it. I What what a fucking? spectacular dumpster fire Really looking forward to the to the rollout of this and this is an article ⁓ open AI had planned to basically release adult mode for ChatGPT, an erotic setting. Dial up your ChatGPT to suck. We would literally. Just sucky, sucky. Oh my God. What was the actual quote from Sam Altman? said, we need to start treating adults like adults. You turn 18 and you're issued a fleshlight and a dildo and a live connection to ChatGPT. You're an adult. Yeah, this is how we treat adults. I don't know. So Sam Altman, who's desperately just looking for money, I think at this point it's really clear Sam Altman is just flopped sweat, looking for money however he can. Hey, guys, there are no bad ideas in brainstorming. I think this went a little bit beyond brainstorming. This went to the point at which Sam Altman, they have an advisory board, ⁓ you know, a show. a show pony advisory board, much like Metta's advisory board where Zuckerberg's just like, thanks guys, I'm gonna do what I wanted to do anyways. Thanks for the rubber stamp that I will ignore. And the advisory board came, so Sam was proposing adult mode, treat adults like adults, and allow them to have sex with their computers, which, that's not really that big a change. We've been having sex with our computers for at least 30 years. And the advisory board was like, absolutely not. ⁓ Just a horrible idea. We already don't have good mechanisms of control for age verification and one of advisors is like, we're basically about to publish a suicide sex bot and something along those lines. And then Sam Altman indeed was like, okay, thanks for the advice. It's just advice. F you. It will be out in three weeks time. Needless to say, there was a ton of blowback and way too much. attention on this story and they've turned tail and decided to cancel chat gpt after hours which i don't know do you have any thoughts on that fuzz you have anything on that i want to no i i honestly just wanted to listen to you talk and then make a bunch of jokes about jokes six jokes well it's also i mean i have so many thoughts about this so one of which is I wanna, I would have loved to have bet, like, I wanna be in the retro for the development team for chat GPT after hours, like. What did we do well this week? Like, I'm a QA. ⁓ I'm a QA engineer at OpenAI. ⁓ cool, man. Like, tell me about your week. It's like, well, I had sex with a computer. It's like, I don't know. ⁓ my god. I mean, because, I mean, the other part is like, is this, I mean, and I don't, I mean, obviously I don't know the details of how they're training these models and whatnot, but is this like, How much of this is, at what point in the training cycle is the sex part? Is this like pre-training? this fine tuning? Is this sexy fine tuning? Like, is this a rag? They give ChatGPD 5.4 a couple mags, put it in a room, a couple dirty mags. I don't have much more on this one other than I'm both. absolutely relieved that it's not going to happen and sorely disappointed because it would have been. Yeah. Yeah. think that I think we could agree that it was a horrible idea and we shouldn't do it. But at the same time, it's damn this would have produced so much so much great content. It's like, OK, OK. All right. I've got a pitch, too, which is it's a coding agent. But as a coding agent with backed by the sex bot, LL. ⁓ Yes. Yes, we could have what what could have been I have I have I want to pause for a moment Ian and Ask you a serious question Are we doing a disservice to the world by generating more? Content like more podcasts one more there are Like are we? the problem Are we part of the problem? question Maybe. Cool. Do you want to do you want to pull one more in here? Well, I real quickly the whole meta acquires mult book. First of all, I had to some explain to me last night what the hell mult book is. You didn't know it. And was like, oh, I'm on my head. I had I had heard I had heard the description of it, which is like it's a it's an AI. What is it? People say it's an it's an agentic social network. And I did it. It's just Facebook for a right. Yeah. So The first time I heard that, that's what I thought. But then I also was like, that can't be what that is. It's got to be something different. that's not, it's just a bunch of posting some shit and you can go. It was this, this for me is just so such a beautiful representation of how terrible a leader Mark Zuckerberg is. You know, you know that it's like metaverse all over again. Kudos to the Molt Book team for just like creating something that a dip shit gabillionaire was like, we need more honestly feels like it felt like, now that I know what it is, I feel like the people who created this, was the total troll. And now they're taking, they're running home with a giant bag. for them. Yes, but this was an interesting one and Malt Book, interesting thing about, not the interesting thing, of like observation of Malt Book generated a tremendous amount of buzz when it was first launched because you had these kind of conversations back and forth and like the agents created their own religion and there's, you know, it was actually more like Reddit for bots than Facebook because there would be kind of like... Yeah, that's what my friend described it as. She was, yeah, she said she was like, it's Reddit. Yeah. Well, so... The things that came out later, a few things. One, within about a week or two, it devolved into spam bots trying to either install malware on your machines or get you to buy crypto. Like really, really deeply. ⁓ God. It We always end at the same place. it's agents or humans, it doesn't matter. It doesn't matter. It's like, follow the dumb money. ⁓ and two, there was actually lots of evidence that quite a few of the posts were not made by agents. they had, okay, and three, the security of the site wasn't great. I'm just gonna throw that out there. ⁓ and four, the way that the site is designed, of course, how do you connect to this thing? You wire up your open claw to it, So of course, everyone's Yes, that's how it that's that's that's the whole thing. knew I knew it was I knew it was related to open claw or molt claw clawed. What was it? What was it originally called? Moldbot. I think it was Moldbot. I feel like it was clawbot. was clawbot. No, no. Then there was Moldbot. they didn't buy the Moldbot was the middle one. Yeah. Well, Moldbot, didn't buy the URL and then like a spammer bought the URL and was installing malware. That's why they had to change it. Yes. Yes. They change it to Moldbot without buying the URL. So idiots. If I've learned one thing in my many years of development is the minute you have an idea, the most important thing for you to do is spend at least two or three days buying a domain for it. Like before you do anything, definitely don't do any work and don't tell anybody about it. Be like, I gotta get the domain just right. Which by the way, I did buy a domain for us, which I don't know if I've told you. It's robot.grape. Yes. I don't know if that's good or not. I thought that was fucking wonderful. I didn't even know there We will, on one of these episodes, we are gonna one-shot our website. it's gonna be Oh, that's a great idea. Yes, I love that idea. It's on the mirror board. Is it? It's on the board. Yeah, I think so. We have great ideas. I don't have a great memory. Over to you, Ian. You were saying something about Moltbook. Oh, right. So I knew that they, so I knew Claude sued them or Ceasen desisted them, and then they went to Moltbot. And then I never read the story of... Moltbot to Openclaw because I just assumed they were like well Moltbot's a dumb name No, no, just we should change it. We changed it in a hurry and now we've got a better idea and we're going to openclaw and I was like, Yeah, that makes sense, but I never looked up the fact that they just fucked up. Yeah, they just messed up. It's no it's no big deal But ⁓ the other so the other piece of that which is just It ties into kind of the larger narrative I think you hear Ian and I talk about, which is the way that Moltbook is implemented, there's actually a heartbeat ⁓ prompt, I guess, that when you, that you prompt your open claw to pull a heartbeat from Moltbook, which is, it's gonna download a file every five minutes or an hour or something, and then run it. And so it's basically like, yeah. So it's like, ⁓ my God. So it's just like, hey, on my machine with total autonomy over my machine, go to like some random website every five minutes and pull down prompt instructions and do what it says. And it's just like, I'm tired boss. Okay. Well, that's probably enough for today. ⁓ I've got to get to work. Well, actually, sorry, the one last thing I know you gotta go. I do want your thoughts as someone who's been in the film industry, because obviously this is a big fucking deal. ⁓ right, right. It the whole Screenwriters Guild strike of yesteryear. Yes. But what do you think about, so that there's been a, I can't remember, what's the director's name? Don't recall. Do not recall. Director Korte Voorhees. What has that guy made before? Anyways, he wrote a film and he really wanted Val Kilmer to play it and Val Kilmer obviously tragically passed away from cancer last year, right? Yeah, but he's been working with the estate of Val Kilmer and they have agreed to allow him to make the film and entirely AI generate him in the role. I don't know how to feel about this. It's fraught. It's incredibly fraught. And I think that, what's my reaction? You won't be able to stop this. Like this will happen. This is going to happen. In terms of the ability of models to generate ⁓ repeatable, identifiable characters who perform, know, repeated, like, that you can have consistency of character and scene and world between shots, basically, Which has been the limiting factor up until now. And that's continued to improve and it will, that will improve to the point at which you can generate stuff that you can, there will be a fully AI generated movie at some point, for sure. I think we'll have to see, will I watch this film? Probably not, sounds like a dumb plot. ⁓ I'm more of an action guy, you know? Put Dead Val Kilmer in Top Gun 15, I'm in, right? Yeah. For me, that's happened with age, unfortunately. Still to this day, my favorite film ever is Eternal Sunshine of the Spotless Mind. ⁓ really? OK. If it came out today, probably wouldn't watch it. ⁓ You're like, meh, sounds dumb. Yeah, it sounds dumb. It's it's too like I don't know. It's just being older any kids. I think that It's too it's too emotionally draining. Mmm, right. Mmm. It sounds tiring. Okay. Yeah, sounds like a thing. It sounds like a thinker I don't really yeah, I think enough. I just want to shut my brain off and see see the superheroes fight it out Yeah, fair enough ⁓ I think I do suspect that there will remain and continue an audience and a place for like real film or real performances, regardless of what you're able to do in AI, I think that there will be something, some sort of, and whether that translates like, whether you can tell the difference or not, like, you, your knowledge of the provenance of this thing will change your experience of it, right? Like, it's, and so, I do think that there will still be a place, but I also like, I just have no idea. Yeah. I wonder, I often wonder what would Val think? Well, we own the rights to the character and you perform the character and now we're going to make the character, you know, do whatever we want. You actor. GPT after hours with Val Kilmer. And on that note, folks, that's the episode. Fahs any... Any last thoughts you'd like to share with people before we push on for the day? No. No last thoughts for me. Just keep calm and carry on. freak. Try to tamp down the freak out, everybody. This is mostly what I just tell myself every day. It's like, don't freak out. Change is happening. If you can carve out some time to **** around, play, play with the systems, experiment, see what it means to you, I think you're gonna be in good shape. So. With that, ⁓ episode five in the can, we will, I will be publishing, yeah, the only other note here is I will be attempting to publish episode four tomorrow. That means that we have crossed the most dangerous threshold for a young podcast. 54 % of all podcasts never make it past episode three. So if I can survive one more day and hit publish. We're in bro! We're locked in! with us, internet! Thanks folks! Have a great afternoon, morning, whatever, wherever you are. That's a wrap! You just survived another episode of Yelling at Robots. Better luck next time.