#16 xAI Exposes Private Data, Nano Banana, Are We in an AI Bubble?

All right, I have a blue check mark and we are live. Yay. Uh, let me That was quick. That was really quick. Hello everybody. We probably will have uh a few new faces today here hopefully. Why is that? Who's whose faces are new? I don't know. like from the from Oh, from your Okay, because we're Yeah, because we're streaming to your Twitter. My Twitter as well. Set up to stream live now. Sketches. Do you know I always get this popup on YouTube Studio whenever I click to sort of go live? It asks me, do I want to go live now or do I want to stream later on? I don't want to do either of those things because we're already live.

Yeah. I have no idea why. It's just every single time I can't just close it and say, "No, no, I don't want to do any of those things." Just bugger off. Bugger off. Streaming software. And then it opens up a new stream if I do that. No. Anyway, let's wait for the audience a little bit to join. I see seven viewers. Uh so guys, just let me know where you're joining us from. We're taking like a little moment until more people will join. Yeah, we're going to talk about a few interesting things. Um, join us on YouTube if uh if you so desire to join in the chat because we don't monitor the chat on Twitter.

Uh, come over to Command AI YouTube. Okay. Um, we have got, let me run through the headlines because we weren't here last week. So, now we're yeah, playing catch up on a few little bits and bobs. Nothing too crazy. Um, yeah, you've probably seen in the title, XAI exposed in 370,000 private Grock chats. Crazy. Um, nano banana, which is not actually you better say it. You say nanobanana. Nano banana. Yeah, I will. Uh, I I genuinely thought that's the official name. Um, but yeah, it was the it was the secret um code name of it whilst they were testing it on uh LM Marina. Uh but officially yeah 2.5 flash image or whatever. It's just you know we'll go over that.

Um I main topic. Yeah, we should go with um like our typical intro or something similar to it. Oh yeah, we haven't thought about that. Yeah, let's uh let's do it. So I'm going add this to our notion and I'll go with the Yeah, I we we got it already. Yeah, it's just updated. There we go. Um, is AI in a bubble? GBT real time. Uh, $100 million framer investment and then producing an AI video from it. Clauding Chrome. Yeah. And then we'll see how much time we got because we got tons to go through. I mean just to let you know Claudin Chrome YouTube AI uh YouTube using AI to secretly alter created videos Grock 2.5 open source Apple and talks with Google to power Siri used for Google search and cozy proposal stewardship of Chrome but I've added them sorry they are added I don't know what you're what you're mentioning in notion if you check ocean.

Yeah. I don't know what you're saying. What are you What are you telling me? Yes, it's in notion. Yeah, these things are in notion. No, like the if you read like the beginning of the document. Okay, fine. Fine. Um, so with that, should we do the official intro? I think we've given um given people enough time to join. 26 people. Very good. Join us over on YouTube if you want to chat. Otherwise, chill out. Listen to us on your uh on your Twitter. Um should we do the intro? Yeah. I'm going to try and do it without uh reading it. Welcome to Command AI, where we press command F on the internet's AI news and still somehow can't find our sanati.

My name is Samuel Gregory and I'm Kabza and we are here to decode the weekly chaos in AI web design and development. This week we have our nate nightmares coming true. XAI exposing private chats, Nano Banana by Google, we have AI bubble might be real and framer raising 100 millions in uh investment and a lot more in this week's episode. if we have time. If we have Yeah, a lot more if we have time. We've got two weeks to catch up on. So, if you uh if you uh you know, not around last week, we didn't go live last week. We both had other commitments, but in the meantime, AI news has just been piling up, baby.

Um but let's let's jump right into the I'll share my screen just to give you something to look at, but the weather I don't know. I feel like this did not get enough eyeballs on this. Um, this is basically Well, I don't know. Do you want to take this away, Kabaza? Yeah. So, even if you have nothing to hide, I almost spilled my drink. I shouldn't get so excited about this news. Sorry. Even if you have nothing to hide, even if you are willing to show your DMs to your spouse or like your WhatsApp chats or whatever, you still don't want to share your chat GPT and, you know, AI chat dark secrets with anyone, let alone with the public.

And XAI just did that. They just allowed a bunch of private chat to be released to the public to be search engine indexable meaning that people could search it and would find these private chats which is a true nightmare coming true. Uh do you want to take on to take it from here and yeah just to just to lay down the kind of technical aspects of it. Basically, you're chatting with Grock, you come up with something. I think all of the AI um tools have this feature actually where you can actually share your conversation with someone else, right? Yes. Unbeknownst to those people who are sharing their conversations by clicking that share button, it's making it public, which I I think it I mean it's the only way this feature can work is by making it public, but it was also allowing it to be indexed.

it didn't probably didn't have like a no index or something like that which by the way Google can choose or not choose to um honor. It can see a no index and still index it but it's you know at the goodness gracious of their heart that they would uh not index it. And these were conversations that include um manufacturing of illicit drugs like fentinyl and methamphetamine constructing a bomb. What are you guys chatting about? God. Um, and a word I'm not going to say just because I know it gets a bit um, it gets flagged on on on YouTube here. Maybe you can see that. But yeah, um, yeah. But amongst all this, probably private chats, probably things that people were talking about that effectively you wouldn't want to be indexed.

So that being the sort of um how you say the the kind of technical aspects, the the thing here is like for me personally like if you think back to Cambridge Analytica and the kind of dawn of social media, I have a video coming out which talks about what people are afraid of AI and a lot of them talk about the security aspect, you know, and I know this is not directly related to the overall security of AI. This seems to be some sort of miscommunication to the user that it would be indexable by AI by Google. Um, it could be a coding kathuffle where they forgot to add that no index command or whatever it is, however this thing got leaked.

It just shows that inevitably we expect more from these huge companies. We expect them to do better. And between you and me, I don't expect XAI to do better because it's owned by Twitter and Elon Musk. And Twitter is an absolute show right now in terms of how good it is as a product. It bugs out all the time. So, I don't expect that much from Elon to be honest. But broadly, we do expect more from these tech companies. We do expect them to have that extra layer of due diligence when it comes to security. But clearly, we're not getting that. And clearly there needs to be some sort of legislation, some sort of, you know, um, things in place to make sure that they are just even checks, you know, which I think I haven't looked into this, but I think that GDPR are treading around this very carefully and I think they might be a a leader in this sort of thing, but again, brings trust into question.

Exactly. It's been on people's mind for sure because I've seen it as a meme uh and I'm sure a lot of people have seen it. The videos where people are asked to share their chat secrets with or like chats with um their closed ones and they just throw out their phones out of the window. Oh, look. You maybe can't see it. Look, maybe it's blocked. Really? Oh, interesting. It's like Oh, no. No, it's all good. There it is. Okay. By the way, can I check your Instagram? Can I check chat to you? No way. No. No way. Yeah. By the way, just this is like a geeky thing. Uh if you share your screen with somebody and you can't, you know, share the video because it blacks it out.

Uh you can bypass it by turning off um hardware acceleration in your Chrome. There you go. What a loser. He doesn't. He just throws the case out there. Just he doesn't have the ball to throw. He doesn't have the balls. Come on now, man. Yeah. Still a lot of people don't put their phones under water still because I don't trust waterproof. And it's like, come on. Shower. Yes. When I tell people that I wash my phone like three times a week, they look at me like I'm messing with them, that I'm like making a joke. I'm like, "No, I'm serious. I put soap on it." Like literally, yeah, take a shower with that. We sing.

I have one of those sticky magnets like on my like I go in a shower with a podcast. I never listen to music in the shower, but with a podcast mostly AI. I should get a life. Yeah, maybe. Maybe. Yeah, you got you got to be you got to be um uh what do you call it? Efficient. You know, we're all busy people. We've got to get it in. I, you know, I listen to this stuff when I'm trying to sleep to try and cram it all in there. Yeah. You know, that's how I learn. I sleep on my books. Like I put them on my try and try try and uh learn them through osmosis.

Yes. Yes. It works every time. Yeah. Anyway, back to privacy and uh AI like we are we are at the cutting edge of of these are the cutting edge technologies and they're letting us down and people are having hu who was it um did did uh Sam Alman even talk about like telling saying to people not to have private conversations with AI. Wasn't that something that happened this week? I don't know. I'm not sure. But honestly, I have a tons of private conversations with AI and a tons of screenshotting and sharing like um words that I don't understand. There are like obviously it's not my native language, English and German, and I just screenshot things in the context if I can't copy it because yeah, sometimes you can't copy it out of a video or something.

And I just paste them in. And for those two types, I use temporary chat. Yeah. And well, yeah, I sometimes utilize that, but sometimes I just completely forget. And you know, at the end of the day, the end of the day, um, if you look into the T's and C's of a lot of these companies, they do talk about not training on your data, right? They do talk about not storing your data. And in fact, um, OpenAI were transparent about the fact that the New York Times were forcing them to keep data because of an ongoing trial which was against their terms of use. And this was a few weeks ago now. So, let's be clear, these companies say they're not keeping your data and all this, that, and the other.

But, you know, what this XAI thing raises into question is one, um, are they, you know, are they even capable of, you know, treating data? Do they have the best intentions when it comes to data? Do they have the technical ability to maintain data? And then secondly is just purely the trust aspect like do we trust these companies to not be storing our data because there could be another Cambridge Analytica thing where it turns around it's like oh yeah sorry we have been storing all of your conversations and and learning off it and stuff like that. So at the end of the day, they're so bad at hiding or storing data, it will probably come out in the in the in the public domain anyway.

So not too sure. Yeah. Speaking of have a company who has a lot of data, Google. Google indeed. They came out with their new model and the great name for it. Well, I I'm sorry to break it to you. It's not quite it's not nano banana. Like that was the code word. I'm looking for it by the way guys. Um let's look at uh Gemini 2.5 flash image. Um, so everybody's talking about Google's nano banana and it's apparently it's not even officially called this but that was their uh internal name and now we are going with it because nano banana is much better than whatever this is Gemini 2.5 flash image. No, thank you. Nano Banana is Yeah, much better.

Yeah. Yeah. So, this is an image model and it's apparently very capable and it can do like really cool edits. And the point like the main selling point here is at least from what I've seen is it can uh remember faces really really well. So you can give it your picture and ask it to to create a new picture of you like in a new environment completely in a new style and it still retains the like your facial features. I think I think it's just in general it's just good at maintaining the aspects of the image that you don't want to edit. You might have seen these images or these videos of people saying uploading an image and saying don't change anything about this image and then doing that about a hundred times and you you slowly see this person like turn into a slug.

So even when you ask it to not change the image, it still does. But as you can see here with the guy with the red t-shirt and the black t-shirt, it's not it hasn't changed anything about his face. You know, arguably they are pretty pretty identical. Exactly. And this goes hand inhand with the VO3, the video model, because people use it to create videos of like the same person with multiple shots, multiple angles, and create like really cool shots with it. But let's talk about this other use case. So this is like truly bananas. Is it how you say it? This is truly bananas. Like imagine banana. Banana. Okay. So this is truly banana. You can draw.

I need to like get this right so we can create like a nice short out of it. That's why I'm like trying to get it right. So, this is really banana. Imagine you could draw a line or an arrow on a map on Google map and then ask ask it to show you that place that from that perspective and in an image and Google just did this with nano banana. So, as you see in the examples, you can draw an arrow and it will tell you what it would look like if you would have a camera there. That is crazy. So, it's not only understanding the context of the image and the question. So, this person is uploading an image.

It's not uploading any text or instruction. They're uploading an image with essentially the instructions in it. And you could assume so much from this, by the way. It's just No, no, you can. You should ask it, by the way. So, you say, "What does the red arrow see?" Oh, okay. There we go. Yeah. Yeah. What does the red arrow see? And it will show you. And this goes into a a discussion we had a few weeks ago with what that uh procedural generational that model that could create like video game type of environments, right? That you could walk in. And now we have that with Google Maps. And at that time I specifically I said like I'm really excited for this existing in Google maps.

Now imagine if we can drop anywhere in a very like it it will look like Google Earth eventually but much more detailed much much much more detailed like 4K quality exploring Earth. That is insane to me and potentially even play games in it. And we are getting there step by steps by these like different models. But yeah, I I think we are still a few years away from like the real time game gameplay, but it's it's really cool to me. Yeah. The other thing is, and you mentioned it briefly, is that it's really good at just maintain I saw it up here or maybe down here. It's it's it's good at maintaining like the style of something.

So, you can see this person here is a bit grainy, kind of like, you know, 1990s. This one's a bit more modern, but it's matching the style and the aesthetic of where you're what you're asking. Um, one some great use cases actually is uh let's try and find it in this Matt Berman video is to restore photos. Um, oh yeah, colorize old black and white photos and things like that. And it does a fantastic job. Maybe it's that that's a good idea. My parents uh didn't have a a wedding video. They had they have photos but they don't have a video. Maybe I can make a video for them. So this is an interesting one as well and this is why we talk about it being a Photoshop killer.

Like we this this prompt is asking to highlight this building and so it's taking it upon upon itself to make this blue glow. Maybe they asked, maybe the prompt was about glue, blue glow highlight, but then also giving you information here. Now, you'll see I I don't know what this is, but I do believe this is not correct or that there's some text here that it's that's particularly bad at, but you you're able to edit photos. This is another example as well. You know, you're able to edit photos and and as we said earlier, maintain the majority of the image. And um that's insane. C can you please go back to the highlights? And I my my brain keeps going into creating this duplicate of Earth that is like highly detailed and highly interactive without the need of objects.

So imagine this is right now is an image and we know we are we are capable of making videos. So soon maybe we will interact with videos in the way that we would interact with Google Earth. Like imagine you could like touch your screen and and tell it to highlight a building and teach you something about it and every essentially like every frame will become a generative frame. Not nothing will be Yeah. It's all fresh. It's all new. There's no Yeah. There's no preconfigured uh way of interacting with these tools. It's all on the fly. Um interesting as well again just this idea of maintaining that given it a white jacket to put on this guy and this image and you wouldn't know you know he's maintained his t-shirt and all this sort of stuff.

So it's just and this is it maintaining the styles the classic Muhammad Ali make it into a Simpsons thing. This is a restorative thing. So just make the colors punch as well. Just again just maintaining what it's and this is this is fantastic. The way to use it and I can't show you right now but the way to use it is if you go on to Gemini or AI Studio, you don't necessarily choose the model. It's not a model. It's it's built into uh Gemini 2.5 flash and you choose image as your kind of output. You know, you can select from a from a list of outputs. Maybe I can get up for you. Maybe yeah, it's getting really good like image um generation.

I am really good. I am logged in. By the way, we have quite a lot of more viewers. We have 73 people from Twitter. Hello everybody again. Like thank you so much for being with us. Um drop a comment by the way. So here it's literally this is it. This is Nano Banana. So now now we can describe an image. I wasn't having uh much what did I say? I was it's not even in my list. Oh because it was on my other account. But um yeah uh flash image nana banana. There you go. So, can can we do uh a Google map test right now? What do you want to do? So, tell me your prompt, baby.

Um go to Google Maps, maybe somewhere that you know, somewhere in London, and just take a screenshot at an arrow. Do you have screenshot X, by the way? Uh not on this computer, unfortunately. So, we're in London. Here we go. Somewhere that we both know, but somebody somewhere that is not probably Tower Bridge. Here we go. Tower Bridge. I'm including myself as if I as if I London Bridge once. Yeah, you're an expert. Well, you were you were around here. You were around. Yeah. Yeah, I Yeah, I spent like the few a few days. Yeah. Uh anyway, yeah. So, we're at Tower Bridge. Should we take a screenshot and see what we can come up with?

Yeah. And we we need an arrow apparently. We need to add an arrow and say, "What does this arrow see?" We can do that. Screenshot. There we go. Boom. Boom. Boom. Um, this is interesting. I I I like doing this. And even Here we go. Yeah, let's do this. See if we can get St. Paul's Ca Cathedral in there, maybe. By the way, we don't see the screenshot editing uh in the live. That's fine. That's fine. As long as we will see the final result. So Sam is right now trying to create a screenshot with an arrow. And so here it is. I've uploaded a screenshot. You can see the arrow there. Yeah. Pointing at tower bridge.

But that's an easy one. But anyway, so let's show me what what we say. Show me what does this arrow see? That was the problem. What does the red arrow see? This is the thing. I tried this this morning because I was trying to design a tattoo, but um it was just failing. So, let's see if design a tattoo with AI. That's bold. Yeah. Uh okay, definitely selected that. Okay, here we go. Let's be let's be maybe we're just taking his prompt too direct. Um generate an image of of what the arrow sees. Here we go. Um yeah, the thing I like about Flash as well as an image generation is in the name. It's very fast at creating images.

I mean that is probably and I can tell you this is probably more like it's probably more like where that is these houses. It's around um rather height than whopping. That's what kind it's not but there you go. That's what are we getting wrong? Can you check the tweet uh the tweet very quickly? I think this is the thing about, you know, people sharing their experiences. They're very choice with their um Yeah, I agree. It's draw what the they're saying draw what the arrow sees here in the Eiffel Tower. Um they probably tried this a few times and didn't get a good result. So maybe we just because what they are getting significantly better. Yes.

Yes. That would be great. I mean the Eiffel Tower also in that example wasn't great but I mean um let's see you can see my screen right generator. Yes the position that we are showing is very easy like I was I was about to say let's go harder at it but I mean this is the Millennium Bridge. No it's still not. There's some pools. It's not quite there. Yeah, it's this is interesting though. It's it's generating a little watermark so you know it's AI. I mean that's easy to take out ironically. You probably take it out with AI but um yeah not quite working for us but you know this is AI. This is how it works.

This is unfortunately what you get sometimes. So um yeah. Yeah. So more playing is necessary I think. But it's it's a good start. Like it has I mean where where my I had just a lot of hopes. I thought it would be like very exact. But let's let's go into the main well not main topic, don't get me wrong, but like the big topic of what we're talking about about AI being in a bubble. Now you shared an uh CB CNBC article here. Yes. Yes, I will get it on the screen so you can talk us through it. Sam Alman. So Sam Alman, the CEO of Open AI thinks that AI uh might be very what what does he say?

I don't want to quote the exact uh title because I don't think the title is very honest. It's very clickbaity. The actual thing is Sam Altman believes that investors are over excited about AI but at the same time AI is the most important thing in a long time. Um, and when we say like overexited, it means that a lot of investors are putting a lot of money in AI. And Sam Alman used the B- word bubble, which is pretty crazy thinking of like he has the the most amount to lose in this case, right? Uh, and he uses the bubble word saying that AI might be a bubble. And this is what we all think because there is a huge amount of money pour into AI and we we don't see a lot of return except for Nvidia.

We can like shortly talk about this a little bit but Open AI they are going to make 20 billion this year but that is like the money that the revenue but that's not profit. They are actually losing money. So they are making 20 billions but we don't know how much I don't know exactly how much they are losing their margins are like negative they are not making any money who's making money uh if you could also show that Nvidia one I mean it doesn't really matter just the point is yeah that great thank you uh Nvidia is making tons of money u mostly from data centers so they are making 46.7 7 billion. Can we can we break this down?

As as a simpleton like me, I'm I'm struggling to follow this. So, okay. So, this is like this chart shows how much money a company is making and what is c what is um making how much money. So, data center time across here is this time beginning of the year, end of year. This is sector. So data center at on the very left we have just data centers all together. All right. And if we go to the all right we see the net profit. So from data centers and gaming and everything else together the net profit is 26 billion. Right. Yeah. and 41 billion of you know 40 billion dollars were from data centers and the end like net profit of the 46 is 26.

So so this includes obviously so the black is the total revenue. Yeah. No, the black is the total, right? And the data center on the very left is just data center. And then we have, this is very funny, gaming being 4.2 billion. It's like I wonder what that means. Does that mean um because like is that is that just their gaming spec because GPUs I'm just sorry I'm mixing my words but GPUs for example are used for AI yet they are primarily used for gaming. So how are they? No, they we have like gaming GPUs and um like H100, H200 like the data center GPUs and they are building these data centers themselves but also selling a lot of these GPUs to these companies.

Yes, there is a very like small part of like the gaming GPUs being used for AI, but that's I would say it's so insignificant in comparison to these big numbers because that's the point like um companies like OpenAI, Microsoft, um companies like goo I I think also Google and um Facebook, a lot of these companies are buying all of these like data centered like these GPUs from Nvidia and they are roughly responsible for I I think like 50% of their profit that is that is crazy and they are all um losing money so the companies who are making the AI they are buying hardware from Nvidia they themselves are losing money but Nvidia is making a lot of money with a lot of like margin so that's that's essentially And that's why we say it AI might be this is one of the reasons not the only but one of the reasons AI might be a bubble because the hardware company the the company that sells the the shovel for the gold mine they are making the money and the the people who are like buying the shovels to get to the gold mine we see they are losing money with open AI losing uh still not making money with 20 billion dollars of profit still losing money and the point is where will this stop?

Will we get to a point where okay we we bought enough GPUs now we are going to make money well there are data showing that that's not the case because the GPUs that they buy uh they depreciate they get they they break down and after a few years they are not as fast and other better GPUs are out there. So it seems like Nvidia will keep selling GPUs and these companies will have to keep buying them but if they don't get to a point where they return the money to the investors like you know in terms of profit it might pop as a bubble because and we we have to see for how long will uh investors just take in these um losses just pouring money into at the end to inv IA directly or indirectly and not seeing any profit.

So AI might be a bubble. I I I sort of you know when I saw this I thought about it basically. I mean they know they invest investors are taking punts you know they are taking punts on what they think is going to succeed and AI is obviously all the hype and and even even Apple have been slapped on the wrist for not being part of this hype train you know and it shows a lack of confidence blah blah blah from stakeholders. I translated this as potentially like a lack of serious education for these investors that they are just throwing their money at anything AI just again in the hope that it pans out for them.

Yes. So maybe maybe what Sam is saying here is that they are in uneducated with what they're throwing their money at and not really maybe they're putting the money in the wrong areas. Maybe Sam is see saying that the where they should be putting their money they shouldn't be putting it in these dumb I don't know photo editing apps or something like that you know or something maybe they're he's just seeing that they're putting their money anywhere and not really thinking about don't know I don't know at the end of the day it's really easy to see Nvidia being a winner because open source man a lot of these like AI tools we've talked about this.

Uh they are getting really like so so much better every day, but at the same time the open source versions are catching up really fast and we might have a day where just on your like gaming GPU or on your Apple computer, we can already run some models, but maybe we will be able to run even better models at no cost to us. And then where is OpenAI in that picture? Why would I pay open AI if I can just run an open-source uh model? Well, I don't know. There is software around it. There are like garden like uh walls that pe these companies are trying to build. You know, OpenAI is releasing a tons of new tools and they are trying to build these walls around like the EOS the this ecosystem that they are building.

So, keep the users in and we keep paying them the money and someday they become profitable, but that day might not come. Exactly. But it's it's interest. It's just an interesting way to phrase it, isn't it, from Sam like saying we're we're simultaneously in a bubble, but also this is very very important. So um yeah, I don't know. Yeah, I have I sort of do feel like we are in a tremendous bubble and I'm hyped for it. Like some people hate it. You're hyped for a bubble. I'm hyped for a bubble. I'm just enjoying the moment, baby. I'm enjoying the journey, you know. You have your popcorn waiting for the bubble to pop. Not necessarily, but I'm just enjoying the hype circle that we're in, you know, and if it doesn't last forever, it doesn't last forever.

At least we enjoyed the journey. Yeah. Well, I honestly I've been thinking about this. I don't know which one is worse, AI being a bubble or not being a bubble. Like if it's not a bubble and as if it's per these like CEOs of AI companies, we will all lose our jobs, that's terrible, right? If it's not a bubble and it's going to be so great that we'll erase every job essentially. That's bad. But if it's a bubble, I don't know, a huge part of the economy is just poof, it's gone. like a lot of people lose a lot of money and I think a lot of things will go downhill like the dotcom bubble but eventually it will recover it will mature.

Do you think we will be affected by the burst of this bubble like I don't mean like obviously we'd have enough to talk about on this podcast if it did have a I'm talking about us as a society. Do you think we would feel the effects of it? Yeah, I think the there there is a lot of I I don't know. I'm not good like enough with economy to know how will that how like the down if it if it turns out to be a bubble a lot of these like S&P 500 companies will lose a lot of money. We'll lose some of our investment that that we have and I don't know how it will affect you know the rest of the economy but I'm pretty sure we will we will feel it.

There will be a lot of layoffs. Um when people have less money, the companies will have less money, there will be more more layoffs, but the technology will eventually um you know catch up to the like it will be on the it will try to mature. It won't be this hype forever because right now as we just saw saw with the nano banana like the examples that I've been seeing on social media they've been crazy but we just tested it and it's not you know the most reliable thing and it's normal. It's it's where the actual state of things are and I think uh these companies are creating so much anxiety for people that that's why I wouldn't mind if it were a bubble because man they are creating anxiety for everyone.

Everyone is like am I going to lose my job and then you go to the tool and you see it's bad. You see the code I showed you the code that my AI was like creating and it's like my my developer would give it like three out of 10. Three out of 10. I think they say three out of five then but no it got worse. The more you spoke it got worse. Yeah. And you know just I I'm you're going to hear me say this over and over again which is ironic because I'm I everything is cyclical and it's like we've we get excited about something new. We try it. We use it. We you know this it's it's it's the rule not the exception.

Is that the phrase? I know the the the phrase exception not the rule. But it maybe we are surprised at the fact that we get super hyped about something everyone AI is all of the things. Maybe this is just what humans are designed to do and like rather than criticize it or um see it as a fault, it's like this is this this is what we do. We get excited about things and we try them out and we we we try and find the place in all of this. And I always talk about this the color the color TV when color first came out color TV we were very we I say we the the society was like I remember it well.

I remember back in the 1920s or no no 1969 was it? No, 1969 people were like, "Why do we need color? Why why do we need It's fine. Black and white is fine. This is just this is society, people. This is how you think that's crazy because you've known no different." But that was you who if you disagree or or against AI or whatever, you would you would have been one of those people who says, "Why do we need color TV?" And I don't mean we don't need color TV, but it it then took films to be very experimental and artistic and creative with color because you you you look at um Wizard of Oz as an example using color in a very creative way.

Um where right in the beginning everyone was just just making everything in color. And while that seems crazy now, at the time it I bet there were people saying they're just they're just making it color for no reason. They're just using it for no reason. And then we start to see creative implementation of color. And again, this just sounds bonkers talking about this because we're talking about a very anti-reate technology and it doesn't quite translate. But this these are the same conversations that were happening. And it's human nature just the AI all the things or to current technology all the things. Everyone wanted an app. I'm going on a rant now. You can tell I'm getting animated.

Everyone wants an app in the app store. You know, everyone's like, "Oh, we we need an app. We need an app." And it's like, "No, you don't. Who's gonna download your frigin stupid I don't know thing like um uh restaurants? Who's gonna download an app specifically for your cafe or your restaurant just so they can see the menu and order their frigin morning coffee when they could easily just walk up? You don't need apps for everything. Same thing, you know, it's just over. It's just cyclical. It's just what we do. So get over it. Yeah. Well, the issue is it's the the hype moment that doesn't die. And all I want is like real world implementations.

Yeah. Like real world use and the ship will steady and we'll find it. We'll find it. You know, apps have now found their place. They don't need to be app store for everything. But there are there are excuses where say excuses there are there are uses where you need hardware acceleration. you need the accelerometer on your phone or you know the cameras or something like that there deserves an app or you need private and whatever you just want that you want to use the secure storage that deserves an app not a menu not a frigin menu for your pub chain or whatever it is you know so we'll find its place and we'll discover and AI will find its place just let people AI all the things and uh be done with it you know just because you don't like it I don't know we're obviously very biased Well, I am I'm more I I think this has turned me into more of an appreciate not appreciator but just a uh more understanding in in terms of the human psychology of it all of like you know rather than getting upset about something or being so anti- something or for something just everyone's just unique everything is different and a lot of things are very cyclical so it's just don't don't get too upset about Anyway, that's my TED talk.

Yeah. Anything to add? Uh, no, not really. I I agree with pretty much everything. It's just the start of it and we'll see how it will settle and we are here, you know, it's a journey. It's a journey. Exactly. Um, we should talk about quickly the most awkward as they all are awkward live stream of all time for GPT real time, which is quite a cool tool essentially, you know, having a conversation uh with an AI and it talking back to you in a tone, in a um accent, in a language that you ask it too with the tonality, the intonation just absolutely spot on. We've already seen this in the app, but now it's available to developers to utilize in their app in the GBT real time API and um go on that prepare for a lot of more a lot more spam calls.

Well, I mean they they used and we see this a lot and it's where chat bots actually first really appeared even before AI. It was just a trilly thing of of customer support and things like that. So yeah, absolutely like see an influx here and it's great to um that chatbt is on it arguably and I have seen comments about the fact that 11 Labs is by far and away better. This was a really interesting part of the conversation where he uploaded a picture of his daughter kind of, you know, playing on a unicorn and looking out the window and he said to he asked it in, you know, am I a bad parent in this photo?

And I love these demonstrations. Google did this where they just, you know, who is following me? That's not someone following you. That's your shadow. using these obscure demonstrations to to kind of retest the logic and understanding of these models and ultimately it said no you're not a bad parent you're just observant but you might want to reconsider it stood on this unicorn it could be a bit of a safety hazard but you're not a bad parent very interesting very very interesting so so so just to understand we already have the voice mode in chat GPT and how developers can have essentially the same tool that they can you use in their own apps. This this is like great news for everybody.

And this is T-Mobile. They've been working on it as a pre-release. And there's a whole shopping experience where you can speak to it. Um, and voice is becoming more, you remember when Siri first came out and people were very like I'm not talking to this thing and you know being very shy about their voice. I still think we should be relatively shy, but you know, no one wants to hear you chat to your chat GBT on on the bus. Do you know what I mean? Like, no one wants to hear I do it in the supermarket, but very quietly like very quietly. There's a lot of noise around you. Show me how I can make situationally.

We are getting more comfortable. So, in this situation, her daughter has just dropped her phone in the water and she's frantically asking T-Mobile, I need a cheap phone. Does it work under my plan? And does it have satellite? you know, you know, for emergencies, does it have satellite u functionality and the AI is able to understand that and walk through that process. So using that natural language where you don't have to type and obviously you're very frantic, the voice a was or the real time sorry AI was able to be implemented into the to the uh T-Mobile app. So yeah, um, voice is becoming far far more popular than it maybe once was. And I think this is a welcome addition, but arguably we do already have some decent solutions out there and this is just a open AI catching up and joining the party as it were as opposed to anything particularly groundbreaking.

It was interesting that um the the intonation, you know, tell me in an excited way, now tell me in a sad way, tell me in this way, whatever. Then I think they asked it to write a poem and change the from um English to Spanish to Japanese about a particular thing, situation or whatever. So, um yeah, those sorts of things are quite nice. Um but but this is like great news in general. Uh I think this goes to the direction of like the same the same conversation these tools will mature and this is just very early like these are for developers right this live was for developers and the first thing that I can think of like we I have a lot of telegram groups that are like with multiple people with multiple languages and when somebody sends a voice message imagine if you like the app would use the API from one of these tools to translate and and use your own voice, clone your voice and then translate it to a different language and uh the listener could use any of the the languages.

That would be for example something really cool. I even think like building something like this for Telegram as a Telegram uh bot would be relatively easy using because I already have a Telegram but using uh CH GPT image uh recognition. Um yeah uh Honor I think that's the phone word. Maybe it's Oppo, I don't know. But they released a phone, a folding phone just the other day, and it has a feature or maybe or maybe it's the Pixel. I don't know. I think it's the Pixel actually, Pixel Fold. And that has a feature where you can have a conversation with someone and if they're in a different language, it will talk to you in your language and then when you respond, it will respond to them in their language.

Standard stuff. What the Pixel will do is actually try and match your voice in real time. match the t the the sound of your voice and it's a bit hit or miss as most of this stuff actually is a bit hit or miss but it's an interesting technology and we are starting to say I don't know what model they're using it's Google probably um probably Gemini you know um and they said they need two seconds of your voice to clone it that's crazy that's crazy that's a bit scary yeah that's really wild um probably doesn't match the tonality and the pace of your lang you know whatever or maybe that's a separate algorithm that runs post training that actually starts to well it just listens to you in real time and tries to replicate that pace of conversation real time.

So we'll we'll see where this go. Yeah, I actually this this made me think of a like an example like an analogy not analogy it's an it's an example. So, you know, we've had Tesla full drive, full self-driving for quite a while now, but it hasn't been a full self-driving, you know, even yet we don't have it. And Elon Musk has been, you know, promising and promising. And whenever they ask him, it's like, well, yeah, we have the last like 5% the the long tail of the the edge cases. And I feel like with a lot of these AI tools, we are still not fixing that last 10%, last 5%. So if it's with voice, like the human interaction still is very clearly human because of like there are very subtle changes and pauses and whatever if it's voice.

Uh with images, we can still pretty like reliably recognize AI images, especially if we take a good look. Uh same with video, even though those two like video and images are getting insanely close. Um but with something like hallucination and this is also about the example that you shared previously with the tech mobile that they they showed like with uh chat GPT. Yes, it sounds good that that person on the phone as an assistant, she or he sounds good. The the words they speak are very clear and it could be any language that they want. But what if they make up a feature about a phone and we know this is an issue. We know that these AI models still very much elucinate, still very much the the data quality that they output is still not reliable.

And yes, I know there are like arguments that well humans are not reliable either. Um, which which is true. I'm not like saying humans are particularly reliable, but these AI tools are still not as reliable as me reading uh a spec sheet of you know a tool or like a document for coding. Um and we we we need to get that and I think this is also what Apple feels is missing. That's why they are not this is one of the reasons they are not truly embracing the game because they see yes there are a lot of nice demos but these are just demos unfortunately I they are not like fully like reliable that's the point they are not fully reliable yet yeah um I just want to give a shout out we are at 94 awesome we're getting close to 100 if you're watching this right now and paying an awful amount of attention, give this a share, give this a retweet, get people on board, and uh let's try and hit 100.

Even if just a couple of you do it, I'm sure we can get a few people on board. Yes. Text your gran. Yeah. Text your text your dad, text your dog, tell him to come and join the podcast because it's a party. Use your other phone. I'm sure you all have multiple devices. Yeah. Yeah. Yeah. Um 94 people. Cool. So we are going back to basics right now. We are gonna chat about framer. Yeah, we have. So this is very interesting. Can we show this video with audio without getting in trouble? Framer if you are watching like we love the video. I I really like this video. I don't think we can. I think maybe actually no legitimate company uses a website builder like framer.

It's not a serious tool. Real companies code websites by hand. Handoffs are good. Actually, if a serious brand used Framer, it was probably a rogue intern. Nothing public facing. Definitely nothing critical. Serious brands you know and ones you'll come to know already use Framer. Mom, can I borrow your mascara? Here you go. So, do you know the clown meme? I don't know the clown meme. So, uh I'm going to share the link. You can also open this. Like framer, well done. Played really well. I love this like idea. Like they are embracing this so well. Um they're taking th this is to me peak of like AI at this moment like creativity. They're taking a very like well-known meme where you have like this progress like be becoming a clown by lying to yourself.

So the text that goes from text one to text four is you piece by piece lying to yourself. And I'm sure there are examples of this. You lie to yourself. Uh but you know that that's the point of it. And they took this and they created a a video ad using Google's AI V3 in one day and it's like actually good. And it's not just me saying good. It's getting traction. A lot of people are watching it and they are using this to show that they are that to announce that they raised a hundred million in funding to take framework to the next level. with all my like I have some like uh beef with framer for a few things.

I think this is like this this is truly great and their marketing tim team team is absolutely killing it. So here's the thing. This is I saw this one. This is good. The thing is this is where you know we talk about AI slop. Okay? That is when someone with a terrible idea with a halfass attitude comes in, types one prompt and releases it to the world. That is by definition AI slop. What framer have just done is demonstrate that with good writing, good good um idea, that's the that's the crux of it. a good idea, you can use AI to create something that's truly impactful. Yes. You know, really that really the fact that this was AI generated is is just for speed.

You know, they probably had an idea. It's like we can time this really well with our launch. AI is the way to do this. You know, they they didn't need AI. Arguably, they could have paid an actor to come in and just do their makeup. That wasn't exact that wasn't anything um groundbreaking where they had some sort of crazy special effects coming in. But it just goes to show that just because you use AI does not immediately label it as slop, right? It's a great idea executed. That's my point. It's it's so good. And if like from our audience, if you are they are typically called like Gen Z um editors or like if you are a Gen Z and have like a sense or like a good sense of humor, you know, memes, I think this is like a great moment to shine.

Like generally speaking, like these companies, they need this. They need this type of creativity because it matches like really good with budget. Like this is extremely low budget. It's extremely fast and it's extremely relatable. Like people watch this and they they get the meme. They get the meme. They are hooked. And this is like a great opportunity if um if we if we have people who are like can connect these things together. this this is a new opportunity because people at framer they are not going to you know I know that they have a division but not every company has people to do these things right um I'm sure they will be they will they will try to hire people who can who who can take this internet culture and just use AI to embrace it and this is what I really like this internet culture moving on and like advancing and not being lost.

This this gets me really excited. Yeah. Do you know this site? Like I have Yeah. I have a question amount of like memes generated with this site that I cannot Oh, I see. No, I've never seen Never seen it before. Image flip. Yeah. I was just going to say I like the fact that they've even got the the makeup the same. The yellow star on the eye, the blue arch. That's what I'm saying. like you see it and you if you know you know that's the and that's the point like framer knows the audience so well I think this is web flow dropped the ball so bad I mean I'm still using web flow and everything but man like the the social presence of web flow is non-existent at this point yeah it's it's pretty dead it's pretty dead but I mean I mean I I referenced Web Flow's tutorials today to a friend um and saying they used to do them really really well because they and and I think the common thread here is that they never used to take themselves seriously and framer are doing the same.

They're literally they're echoing what people say about framer. It's not for serious websites and uh it can't start hold up in production and this and that and whatever. And they're just saying look if we're it's that classic thing of like if we're saying it you've got no ammunition to even attack us. Exactly. all stemming from the fact they don't take themselves seriously and all of a sudden web flow are taking themselves very seriously which you know it you know I we know the people who work there you know that we they're all yeah they are like close friends uh like amazing people like every time I talk to anybody from web flow it's been like amazing talking to them I yeah they are amazing people but it's about the general direction of the company Um they they had this like S tier standard like Webflow University uh back in 2009 2019 22 uh even 2021 it was the like absolute standard there.

You cannot show me any other company who had this level of production like they had a freelancing uh course that was shot completely like cinematically. It was insane like with After Effects and animations and people who were in those videos they all left Web Flow which is very normal you know companies grow and people also like change job but Web Flow couldn't maintain it. That that's the sad part. Couldn't couldn't maintain the structure or didn't want to. Um yeah. So the the the people who were doing the education left uh I know the the person who was doing the the the 3D the vis visualization she left and she's now working at linear which is you know another amazing company right and we see framer embracing AI and the internet culture and they are taking the views and they they are building the next you know group of audience they're building the the community and what do you think like people will use for their website framer more and more be why because that's that's becoming more of a stand yeah relatable and people know it and you know there is no that's the point are there many new people getting like hearing web flow for the first time no there are like many more people hearing framer for the first time so yeah for Um, let's move on.

We we're at 98 and we we are two people away from 100. So, just one of you just one of you needs to share this and we'll we'll do it. We'll do it. Um, let's talk about Claude in Chrome. Where am I? Oh, yeah. This this this is an interesting one. I'm still not finishing like moving to DIA and now you're telling me DIA is dead which by the way has turned like paid and yeah whatever it's turn paid is it? Yes. Well I you can still use it for free but uh I don't know what I did. I just asked something and it was like asking for money. I was like hell no. I'm not gonna pay you yet.

I'm not even close. Well, I wouldn't speak too soon because this is part of the Max plan, but ultimately have released a Chrome extension to give you clawed inside of Chrome or any sort of Chromium fork for that matter. And as you can see from the video here, pops out. Not only can you ask it to do the obvious things that you expect to to come from an AI browser, and if you're not involved in an AI browser, I would I'd probably strongly recommend it. there's enough is pretty it's pretty good at the moment. Um you know obviously asking to summarize pages this that and the other but it can also actually take control of your browser which again talking about DR that's something they've seemingly failed to bring to the to the equation.

They've spoke about it but they've not implemented it. I use comet which does it. Claude now does it which well I say that it's releasing it to a preview of a thousand users. I've requested access so I can give my insights maybe next week, but we'll see. And it looks kind of cool. And um I mean we get into the boring stuff in a bit, but I'm I'm hyped like and I I bet this will be it will be necessary to have a clawed plan to use this. And one.3 million views on that video. That that's quite crazy. and they are just giving like a thousand users access. Oh yeah. Yeah. Um you during the sign up process, you had to give them an idea of what you're going to be building or what your job is and things like that.

So I'm hoping that me as a developer might get a bit of a bump in. You are a developer influencer. Developer influencer, which um yeah, you might be seeing more of my face pretty soon uh with a few AI products. So I'll speak more about that later on. But um yeah, so maybe maybe and and I think that it would be interesting to see how developers use this, specifically developers use this because of course Claude is a very developer focused platform. That would be my one criticism is that actually as a chatbot, Claude isn't necessarily my go-to when it comes to just chatting. I do strictly use it for you know would you call it STEM like scientific research code coding program.

Do you use like the website or do you use it like through cloud code? I use it through No, I never go on the website. I have the app downloaded. I'll either use it through warp which by the way is getting I'm getting loads of referrals from warp. People are loving it. Um can you show it on the screen please? like I I know warp but I like let's see if I can do I don't use it without showing uh the wrong thing uh no war so you're telling me you don't want to share your private chat publicly well no I mean like it won't you won't see I'll show you I'll show you know I'm just joking about you know XAI just doing that uh stop yeah exactly right u I'm really bad.

I better have shared a whole ton of stuff that should not be shared. Yeah, I just have no shame. So, I mean, it's a it's a terminal window. I've got three terminal things here. Um, and I mean, this is clawed. So, if I exit out of that and whatever, I'm back into terminal here. But I can if I go into like auto mode and this is I'm using light at the moment, but I can choose claw. Can you see that? It's quite small. Yes. But I can choose a model here. And actually GPT5 has been great for writing tests and and just really going through a database. And this is code base. And this is warp.

And I can say and the thing is because this is a terminal, this is t this is clued into the um the operating system. So what I can tell it to open an app. I can tell it to what I've been what I really used it for is create. I didn't know how to frigin do this, but every time when before we come live, I type stream here. And this will this is a shortcut to close all my windows and open up the various windows that I need to to create this stream and it opens them all out in front of me and this that and the other. And I said to I just said to it in plain English, create me a shortcut that whenever I type stream, it does this, this, this, and this.

And it just did it because again, it's got access to the operating system. that uses claw to do so. So, um, but it's also a terminal. Like, uh, you might be able to see that, but I'm switching between agent. Yeah, I'm sold. Say no more. I'll give you a referral. Um, this is insane. What? Well, I I know warp, but I just never tried it. I just know it. Just know their pretty website to be honest. Yeah. Oh, it's a pretty website, is it? Yeah, they have a pretty website. Uh, and I like their Lego logo. their logo. I've just put I've just put a link into I love it. Yeah. I mean, it's it's I pay for the year, you know, so I don't really know what my referrals get me, but I pay for the year.

So, you can use warp to interact with your operating system and chat with AI and do whatever you want. Yeah, this is the this is super Yeah, this is interesting. This is super close to the idea of controlling everything like from just one place. Yeah. No, exactly. This is what I've been saying. It's like it's it's as good as having Apple intelligence on my computer because Yeah. realistically because it has, you know, access to the the the computer, you can in theory tell it to write you an email, you know. I haven't done that. But you can tell to write you an email, you know, or copy this bit of text and whatever because it has just fundamental access to the operating system under the hood.

So, it's pretty cool. Anyway, uh what were we even talking about? Claude Claude as a thing. So, the only only additional thing to um if I share my screen again um addition Oh, we we got there. 100 viewers. Well done. You, the audience, did this. Well, I'm seeing 95 on my screen now. It's 99 again. Uh yeah, we got to go to YouTube. So, um, but yeah, the only thing to to mention about the claud here is that what they're very conscious of, and they haven't necessarily showed evidence of it, but they're talking about prompt injection, this term prompt injection, which is something we posited a few um, probably months ago now on the show, where you could potentially write prompts inside of the HTML of your code in order to, you know, do something with the with the AI agent.

Now, again, I don't think they've actually spotted anything or they're probably not going to tell you how to do it, but this is this is something of main concern, which is why they're throwing it out there in stages and just trying to understand it. Um, there you go. Max plan users can join the weight list. So, you can't even join the weight list unless you've got a max plan anyway. So, this is an interesting thing. Obviously, you know, again, I use um Comet, you use DIA. These are all AI browsers. Maybe they haven't thought about this. Maybe they are thinking about it and just not really putting the the sort of their foot forward in terms of this is what they're concerned about or interested in or being cautious about, but it's interesting nonetheless.

Maybe prompt injection is the is going to be the new the new term. Yeah, this is quite scary and still not figured out. So, don't just paste uh any prompts, especially not into a terminal. Yeah, right. Yeah. Yeah. Well, I will, but you you know what it does. Well, yeah. So far so good, but obviously Yeah. um you I mean there's this idea of only it only doing what you ask it to do and it will ask you to whether it can do it and stuff like this. So so far so good for me but yeah as with everything it should be just in general just don't be dumb that should be the most don't be dumb this is great advice.

Yeah yeah this is amazing advice. Don't be dumb. And we are going back to prompting, right? You're you're prompting like after the the show, you're going to do a lot of prompting. Yeah. Yeah. Same. I'm writing some GSAP animations and as I told you, like for just a few simple like animations, I have 350 lines of code and I need to clean that up. Yeah. Yeah. And like I say, I I was saying to you earlier like um AI is only going to get you so far. And I think that what was the analogy that I made? It was it's turning from being just the person just doing everything for you to now it kind of doing the things that you want it to do rather than like specific like make this do make this function do this as opposed to that.

you know, line by line sort of prompting as opposed to just broad fix everything for me because I I've been I've been working on a feature on a on a an app that I'm building at the moment and it's I've been banging my head against the wall potentially up to three days just not getting anywhere with it. And you know what? I woke up one day and I was like, delete, delete everything. Not everything, but delete this branch that I'm working with. Delete the database. Let's start again. And I got it done within a few hours like and it was and I did it plus way more. I was I was implementing multiffactor authentication. So whilst in the first instance I was I said just do multiffactor authentication for organization for users for this that get everything potentially even like PRD level instructions to then and this is all necessary.

I'm never, you know, as a as a developer, this is something I think you come to expect. It's like it's fine. Like it's not wasted effort. And that was my motivation through this where I was really trying to just be patient to work through it because at the end of it there will be learning and I can use that learning to do better next time. So that was the thing keeping me patient whilst I was you know toying the the walking the line of uh shooting myself in the face or you know uh keeping on going. Anyway, it was all done and and now it's even better than than what I did what I was doing before.

So it was just a big learning and and I was then more well informed about how multi multiffactor authentication worked to be able to instruct my AI to do specific things as opposed to figure all of this out for me, you know. Yeah. Well, the the learning that comes with um like doing development with AI is amazing and I I personally I I love it. whenever I start like prompting and you know build stuff I ask what does it do explain along the way and you know figuring that out is quite nice I um I'll show you quickly I released a short which done has done pretty well fairly well actually um where if I open up let me share warp again um I'm loving this feature Yeah.

Warp share. So if I open cloud, um, if you type model, you've got a new one here down the bottom called Opus plan mode. So I've got that selected. And when you go to plan mode, hopefully you can see all this. If I go into plan mode, this is going to use the big boy. This is going to use um, Opus 4.1 to, you know, to to work hard. And this is strictly in player mode. And I really get it to um you know really think about what it's about to implement before then switching or automatically switch to sonic when it actually comes to coding it. And one one bit of advice is because these are hybrid models.

If you say think hard about this, you can do this in chat GBT5 as well. Think hard about this. It will turn from um just a heavy duty model to a like a thinking a reasoning model. And you'll see it in the you'll see it in the logs. It will say thinking whereas it doesn't normally do that. So if you say think hard and I won't always do that because obviously it's going to consume a lot more more of my tokens and not that tokens are a thing in max mode but the point is it will consume up more of my opus allocation. I won't always use think hard but if I'm really kind of like I need you to in the first planning stages of a of a of a you know a thing I'll tell to think hard.

So yeah, that's my that's my cord advice that I found this week. That that's good to know. Like in my cursor, I just checked I have GPT5 max on. Yeah. Yeah. GPT5. So if you tell it to think, I wonder if you'll get or think hard. Um I wonder if you'll get a different interface or whether you'll get different results. But again, GPT5 is is got that automatic um what do they call it? uh automatic routting where we try and decide whether it should go light or heavy and this this will be potentially something you incorporate into your prompt to say look I just need a quick result here do this for me so as so as because you're like you have so much more experience with this specific part like you're a developer and you have experience with like AI coding a lot is it dumb for me to to tell her to cut down the code.

Uh like it I see like it's I asked it to refactor it, but it does it doesn't look like clean code to me. So So walk me through that. Walk me through that prompt. What did you say to it? No, that that's my question. Like is it dumb to just basically say refactor it and make sure that you use less amount of like code because I just did that and I had 300 lines of code and it did thinking for about two minutes and like implementation everything and it returned 350 lines of code 50 more lines. So yeah, I would say it's a dumb way to work in that you are you're not taking charge of the AI.

how I would approach it. I would say give me um look at this file and give me um some ideas on how we can optimize this code. Don't tell it to don't do anything. Don't change anything. Just give me ways I can optimize this code. Yeah. And then it might give you sort of, you know, a few ways you can optimize it and and and then and then choose one of those to go with. Don't just say do this. choose one of those to go or maybe there's one thing you want to dig a little bit more into and say um give me three more solutions on this function of how we might you know optimize it.

get it to almost not gaslight but convince itself that it hasn't looked into it well enough and and say like yes exactly and but but for that they need to know what they are doing by the way right like they want they need the code does yes and no what does this code do I it it will do that automatically it will say first I need it will give itself a task list so first I'll need to understand what this code does. Secondly, I'll look for any maybe type errors or something like this, you know. Um, presumably you're not working in Typescript, but this is just an example. It will try and understand the file um to to give itself context, but it's not, you know, you might include that in the prompt.

Why not why not include that the prompt? But there could be there's two ways to refactor. one from a kind of again understanding what the code does and trying to rethink how it's structured or you could just be using an inefficient you know you could be writing a for loop where really you could use map you know or something like this you know there there are many there there are technical ways or or ways to write code that will make it shorter or more efficient. So yeah, maybe maybe that is too too prompt in itself. Just say show me some improvements we can make. Maybe some antiquated uh coding patterns that are being used here.

Show me some antiquated coding patterns that were being used here. Don't fix anything. Don't do anything. Just show me and then start digging into those, you know. I see. So it's basically using the chat mode first. Uh going into do those details like looking at my code. Um, yeah, I see a lot of ifels and I'm not sure if I'm not convinced that if that's like the needed way. Like quite a lot of ifels for math. And like in my head like some randomized generated animation doesn't need a tons of if else. Certainly not six about like 60 lines of if else. It would need just math, you know. Um, is there any reason why we we couldn't like next week like go live and just go through this code?

Is there is there anything you're hiding or like you know couldn't do for a client this specific one but I can but I probably can make because I I want this to be a component and we had a cleaner le more lean version of this code for a simple version and this is for a client it's much more complex. Uh so I try to fix it with like a teammate and with AI and by the way at the end I will give this code to my senior developer and he will clean it up and we will make a component out of it. But for now I'll try to do that with yeah try another way is to scrap the code and just start from fresh because you you've now learned a lot more about the code and about the functionality of the tool.

No doubt that there might be improvements you've made through this journey and it does have a habit of leaving in old bits of code. You know, I've just had to clean literally my last prompt in cord was to like remove some functions because I added streaming to the AI which meant we didn't need functions anymore that just returned a returned a solution uh returned an answer. So, but it had left those functions in. So, I was like, can you just clean up anything to do with AI and not streaming? Yeah, sometimes it's worth starting again. Yeah, I I agree. I agree. Like probably it's better for me to ask it to strip down the code to like the most init essential functions and objectives of those functions and what the code should do and write maybe even ask it to write a prompt like a prompt for it and then I edit that prompt and create a new new code.

Yeah. Yeah. get it to think about something for you rather than just go out and write the code like do that to think. Yeah. Yeah. Cool. Let's uh let's wrap it up there. I've really enjoyed we we're trying a new format. We're trying to trim the fat a little bit. Make it uh make the show a little bit more, you know, chatty and whatever. Um still a work in progress, but let us know if you notice a difference. Let us know if you think we can improve in places. Maybe I need to just stop talking. Yeah, just let us know. Yeah. Well, uh it was awesome uh to see so many people hanging out with us.

Uh we are at 110 people. That that is Yeah, that is our peak by the way. This is like uh yeah, this is awesome. Cool. Like, subscribe if you haven't already. And until next time, keep on vibing. Peace out.