EP85: Is AI Slowing Down?
01:13:29 Share

This week everyone is talking about an A I slowdown. In fact, the information had a ton of articles all about this. One of those articles said following OpenAI, google changes tag to overcome A I slowdown. We also had A A potential pivot from uh google and OpenAI to take on the enterprise search at glen, which allows you to you make sense of all the junk that your enterprises is is sharing around.

And then we had this you know predictive um tweet from the founder of mixed panel and playground A A I that said my little prediction of twenty twenty five years A I scaling is platoon plateauing steady amazing progress like micro processes versus jumps will begin to see more of a focus on product. So is IT all over? Is AI slowing down? Is progress slowing down to we all need to now pivot to enterprise search s.

well, firstly, that tweet is just such an obvious take. It's like, oh, I actually don't think that things will infinitely scale and that at some point someone's actually got to use the technology for something like, that's my cheeky little prediction you know like, what a bunch of nonsense of course it's not gonna just keep exploding in in popularity IT seems like these people are just news junkies and they're just desperate for like, oh, open in eyes.

Decided the GPT four is so last year and now they're released GPT six, which can like tie issue like this for you and go to school. It's like, yeah okay, it's not gonna keep exploding in in growth. But as we've said many times, we haven't even use the models we already have. There's so much in them and so much they can do. Let's go do that instead of expecting an announcement every week.

Yeah, IT does seem like this. Just so much more work to be done in using the tools we have. And we did predict, I think, late last year. And i'll have to like pull up a clip. I'm not going to do IT because i'm lazy and editing.

I might one day, but you know, you Normally would pull up a clip to say that last year, we sort of did that IT would be now about going and building the tooling around this stuff. And I I think that's not a unique prediction biology, but it's certainly what needs to happen. And I think in the past week since we released the workspace computer, we've kind of seen that like just building tooling and making this successful to people and then seeing how they use IT starts to give you a ton of ideas. Okay, well, you know what? If we made this moral tony ious, and they could actually do real toss in a work?

This is with an existing model with slight modifications, remain tropic to suit the purpose. But you've got to the model sitting there that can be fine tune for any purpose like that, that is around vision or around text or whatever. So the the underlying technologies is there.

People just need to take the time to build absolute definitely saying that it's no unique prediction to say that, that's needed. But to call IT a slowdown is kind of silly because I think that as we've repeatedly said, people are out their building real useful things with that right now. It's not slowing down.

It's just there needs to be some work time. I do not get announcement all the time. Yeah.

I actually see it's speeding up next year as people start to settle down a little bit and i'm sure the'd be more advancements along the way that that well, as I obviously, i'm like just all in now on the the self driving computer concept at scale, I think that I honestly think it's just gona change everything. But I do think next year is, is that maybe th Epace d oes h old d own a l ittle b it.

IT gives people time to catch up and see IT and think through ideas in specific industries and build really amazing applications that can solve like very specific problems um at a ton of scale. So I think we'll actually start to get along the productivity gains next year. But IT is going to to mean that most of the money now is probably not like the software layer, is like the next layer where people will make money, not necessarily the infrastructure. Like maybe it's time to start shorting a the video in video.

yeah well, I mean, it's all GTA run somewhere. Doesn't IT like the the the the models on going away and they're going to be underlying whatever technology gy is built. The questioning is, are people going to be running themselves or are they gna be running IT? But I would argue that cloud provide as benefit most because all this computing is onna happen in the cloud in some capacity, and people are just gonna have demand for GPU and CPU all over the place.

Yeah I think that there's no doubt, especially if you want to if you think through the computer, use slash work space computer concept, if you're in the enterprise and you eventually want to spin up like a thousand of these things to do very specific workload, which I think, I mean, I clearly see that happening now because we're like building IT.

But yeah, you can totally see that they going to want to spin these things up and it's actually gonna a net win for the cloud providers there because they're running all these talks thing on the consumer end. And this sort of leads into what OpenAI announce with ChatGPT. Um not that long ago where they're now allowing you to, with the desktop apps, actually really use the accessibility settings.

I think this is in a bata right now on the map to be able to see what you're saying. I guess it's similar to what we've done with screen sharing. The difference is with the mac accessibility settings of bringing up on the screen.

Now for those watching, you can IT has IT like IT has the ability of your highlights stuff to know like what you're focusing on. And it's not looking at the image is using the accessibility settings to basically get the road text and the road descriptions from the user in the face of what's being displayed. So this I think modern benefits that it's cheaper.

You don't not having to like to the image queries all the time and you just getting the raw text, which the model obviously loves. Yeah I think in that case, it's interesting because IT got me thinking this week with the introduction of the new queen coating model, which we can talk about a little bit later. I use that model and a lot of people have got that model running locally on their mac. And it's super fast, super capable, I would say, like very close to the level we've seen from salad. And my thinking around this ChatGPT desktop application right now is okay. Well, what's to stop if we're really hitting this slowdown in this war with these where they basically it's about how you tune them and how you make them reflect now to extract more value out of them? What's to stop microsoft and apple eventually having in a year from now their own model, try and run that model in their own cloud and then just, you know, obviously, they're going to own their own Operating system for this stuff.

right? Yeah or on the device itself. I mean, that makes the most sense out of all for IT to be a specialized model that runs on the machine of already seen the microsoft pipe days like they all have gp use in the IT would make and then even the video has their own model that runs on device that would make sense to have a local model that can see the screen, can Operate all the accessibility of, you know, interact with the windows or mac API directly, which IT will, in the case of the the actual Operating system providers, and run the computer that way.

Because dealing with screen shots all the time, as you say, it's like a sort of low fidelity way of communicating high band with. Even though it's low quality um it's still high band with. And then you've got these delays between every single step.

If you think about IT, because it's take the screen shot, influence the screen shot, decide what to do, do that thing, take another screen shot, check if that thing was done. And if you run IT too quickly, you're gonna a miss stuff. And if you run IT too slowly, well, it's slow.

And so I think that having an on device thing that can know immediately or event driven, even like when this event happens, I now know that this text is there. I don't need to wait two seconds to confirm IT. I've just got IT.

So it's a much Better paradise, I think, for the running your own P C thing. It's just my opinion is the running your own PC with an A I isn't how the future is going to be. Like I don't wanna sit my desk and watch the A I Operating my computer. Like that doesn't interest me because it's my computer I want to use IT. Um if i'm going to get an A I to go do task to me, it's gonna its own computer and its own account and it's going to go and do the thing I ask for IT in its own environment and that can be a specialized environment that actually set up to suit its way of working rather than pretending its me and having to use like the accessibility settings to try and work out what text i'm talking .

about yeah I can see. I think there's two angles to look at this. The first angle for me would be IT is really helpful to have like you like microsoft titan and I guess now similar this thing where it's built into the desk top. I can I like if I mean A V S code, i'm writing a document, I can spill IT into that and say, hey, highlight this section and say, hey, i'm working on this.

Can you just quickly edit? And I think we talked about this a while back where we said originally we thought this would just be built into everything right, which is kind of is now what we found ourselves going back to sort of our mothership agent like our best for enda system that we're using all the time as that place to sort of access AI. And I think that for this in in this context, I can totally see that assistance on your computer helping and occasionally taking over, like I am working in photoshop and I don't know how to remove the background.

Can you just do IT for me? And then you watch and learn, I think, is a learning tool. It's phenomenal. And you can imagine the voice interaction if they eventually bring IT as well, like I like literally just talking instead of typing IT, which feels probably more natural in that case. So I think there's that part of that which makes sense to me and will will be pretty amazing to educate people on how to do things on a machine. But then I think what were talking about and what excites us even more is that true and work of this thinking about to golf and like act like an employee.

Yeah as in like I was just thinking that like you, your PC up to a point because of your interface into all of the system you use in your work, in your life, whatever IT happens to be. But at some point, the task, if they can be completed in full, the way employee would complete them, you don't need to see the process like you don't need to sit there watching them like I don't I don't hire someone to do something and then sit and watch them do their job, right? Like I say, go to this, come back to me when you're done and then I communicate to me through chat or email or your presentation or whatever that happens to be a video called something like that.

And I just can see, even in the near term, the idea of having these A I stewards who are there waiting to do toss, you give them a toss like you've suggested via like email them or you somehow initiate the task with them. Then they get back to you when it's done, you can give feedback and all that sort of stuff. And I think it's a far more valuable thing for the A.

I like, let's say, just on a cost spaces. Like i'd rather pay for that, then I would have IT sitting their Operating photoshop for me like. And I know if they're not, it's not an apples to apples comparison. But I guess what i'm saying is in terms of my excitement around the use of the computer tool, use stuff, the excitement for me comes with this idea of being able to a synchro, sly, complete, difficult or time consuming task, not sit there watching IT help me out in the moment like a sophisticated auto complete.

yeah. I I would agree with that. I think that the the value, the the value proposition is completely different. On one hand, it's about helping you in the moment, helping you with a problem or figure something out. But I think that is just going to be table stakes like you know, apple are going to just build that into the OS level and and that they're going to build a model at some point or require a model at some point, probably know when this whole bubble acquire a acquire a whole model to do that and micro soft to get to do the same to me like and .

IT makes sense, right? Like for apple, having an OpenAI A PP t hat c an d o s tuff f or y ou j ust s eems s o f utile b ecause f irst o f a ll, you, it's never going to work on an iphone without control of the Operating system because the apps of sandbox. So it's not I can only Operate within the one APP.

So straight away, the entire phone market is not accessible to OpenAI in terms of automating the device, right? Um android, yeah okay, it's possible. But I would there be similar limitations in terms of permissions and things like that desktop yeah but like let's face IT, it's a hack right now.

Using accessibility stuff is a hack and really that I just seems like to quit Jeffery y hinton activities. They're just trying to stay relevant. They're trying to have something to announce. Like you can't help but wonder if the whole A I slowdown thing is just leaks from OpenAI just being like, hey, guys, don't expect anything big for a while, but we do want funding and we do want to stay in the .

media yeah I mean, IT doesn't IT doesn't really make sense to me like there's clearly more gains in these models and and I like we saw IT from sona, like they literally just tuned IT Better the common news cases that people were using. And all of a sudden IT became like this defect. Those sort of like, in my opinion, still oh the best model. Like I even look at their announcement today, like, oh, cool. I can have a sub pod GPT for giving me suggestions, you know, in an accessibility feature which can only see what's rended on the screen similar to a screenshot I didn't announce editing, so I can replace the code a why not like why not allow IT to insure clearly IT doesn't work yet like it's it's having problems that otherwise you would ship IT with that feature like there's .

no reason not to yeah exactly IT IT definitely just seems like some sort of hacky novelty thing that let's face IT. Hundreds of people have implemented themselves. Like when anthropic released the two years straight away, everyone wrote local programs that could control their local computer to try quickly realized these limitations to IT.

And it's really just an experimental trial phase. I really do think IT comes back to OpenAI not knowing what they are as a company like they don't really know. And you said this this morning, pride of the podcast. I don't know who they're selling to. I think I don't think they know who their .

audiences but sahl before we sort of trash talk that tweet or you did, but he doesn't a really good point just coming back to IT. He says the A I labs will have to cut, burn.

So the money they're spending for those who are unfamiliar with that term and finally ask themselves, who is this fall? Why do they want IT? And I think this is sort of a displays into that point of, you know, are people really going to keep paying a subscription, but something that might become table stakes on a mac in a year. So I to me, that is a dead end for them as a business. Like if i'm just working at like i'm in charge of a sudden day business, total dead end.

That's right. Like even if I am wrong and Operating your own computer is the hottest thing ever and everybody just wants that. Like you say, there's just literally no way you're onna compete with the platform provides they provide the platform. Like if IT build into windows are built into mac O S, what can you do? Like you just not gonna have a third party outcompete them on that.

Yeah and so then I think about, okay, like let's you there's also rumors that google and OpenAI are working on sort of their own computer. Use A P, I, similar to what we've got from anthropic right now. And I mean, that gets me really excited.

I think that's like, you know obviously not hitting a wall or slowing down on those technologies are coming affiliation. But you just you think about that, okay, what's google's advantage? Google advantage.

Is crime like that their advantage? You know the majority of people use the chrome browse of whether you know people will admit IT or not. It's just fact. And so you know they have controlled that's their platform technically even though it's the wear and it's open, I mean, that becomes their platform.

So you would think with with some sort of computer use of germany, I you would focus on their advantage is the browser and Operate things like they've got the j way, which we are own workspace computer on symptoms this way like we've been using people have been using extensively, like logging into either an account they've created just for their agent or their own account and getting IT to write dark sculpin research, put put those things in documents, create sid presentations. It's terrible google sheet and excel right now, as i've seen through my experiences throughout the week. But you know it's gonna a get Better.

And so to me, that's their advantage is like controlling their things. They've got hook s into like hotels and fly. And um you know they're probably best position to create the most practical thing, which is like a chrome process in the background, going often doing things for you in the Brown .

yeah and given the ninety percent of the work, I mean, sorry, i'm not just gonna make up start, but like a lot of people do a lot of stuff in the browse up. So if you had the perfect tool that could just completely Operate the web browser, you're a long way towards the goal of computer automation. And then the other stuff is just bone us if you can get IT. But if you had a chrome now that had a german I in IT that could order IT like do task on command, that would be A A massively popular feature.

I think. okay. So then that brings me back to the west thing ever.

why? If you're OpenAI, are you not focused on the web browser? It's open, it's accessible.

No one controlled that. Apple doesn't control IT. I don't control the web. Microsoft don't control the web. You can build chrome plugins that work in edge, that work in maybe if you wake IT a bit, uh in crime and you can control a tab so you could have a computer going off in a tab in doing task few quite easily. So I just make I don't get the strategy and maybe we just don't have the whole pitcher yet like maybe this Operator thing comes out and IT does control the way browse or whatever. Um but I think it's just as we've seen from even rolling out work space computer like IT, the technologies is just not at a point yet without a lot of a tools built around IT where it's that effective like IT IT sometimes IT absolutely blows you a way at the point you like this is the future amazing and then other times like, man, this is so dumb.

Yeah exactly. I think that's the thing. I just get the impression that everyone suddenly realized how we can do this. Like this is exciting. Like let's try IT and that LED to different people making a lays pronouncements about what the future is going to look like.

So let's let's talk about like I I I think this week we released on cemetery, the work space computer and IT IT is exceeded our expectations in terms of how many people are interested in and just the exact and level for IT. Um you know we really didn't anticipate how many of you would want to use that and tried and literally been just like on the line constantly to ourselves from begging them to give us more capacity. So um I I want to talk about IT and share some use cases.

But I also thought that would be really interesting to talk about is just uh, the feeling that you get from using this kind of thing for the first time, like I think I mentioned to you in the wage, that IT sort of is like the first time I ever used a baby s or the internet um that like the first ever internet like and someone on discount actually said this is like hearing the dialog tone IT IT was actually a beautiful quote. I'll try and find that in a minute uh and and read IT out but IT I mean i'll have to read IT to do justice IT was just such a beautiful quote about the feeling of of doing IT for the first time. So obviously, you haven't had that much trying to play with that because you've just been trying to get IT to stay on line. But but do you .

have that similar .

feeling? I certainly do of like this is something different.

I think when you first see IT working and actually accomplishing a task. Yes, it's just, I guess, is the possibilities like you think about, wow, what could I do if we can do this? And I agree, it's that magic like you get that is why we do this is why we started the podcasts, is why we do seem theory.

I think and I just for me, it's how unexpected IT is like when we talked about IT last way go a week before or whenever IT was, I thought IT would be like a fun experiment we did for the podcast just to get to know what the technologies is capable of. What I didn't expect is that there would just be such an overwhelming interesting in IT, like a really cane interest, like people want to see the capabilities, people have ideas of how to use IT. Like this isn't just a toy.

Like, yeah okay, the technology is very infant and makes a lot of basic mistakes. Um IT falls apart easily like there's a lot of problems, but we can in our podcast mindset, put those problems aside and see that all of that will be solved like we know that all of that will be solved and therefore, you see people and we're not talking about these aren't just programmer and technical people. These are people in all sorts of industries who straight a way like i've got ideas for this.

I've got four or five things I want to try and uh uh of using the computer use for that. So this clearly, there's a lightened for IT, and there's also, I guess, just an inherent understanding of what this cape of is. And I think that's why it's so exciting because it's completely uncharted territory. And there's really a lot of ways you can go with that. Like i've been saying this, i've got pages and pages of writing in my new book here of ideas.

I want to try to make the thing Better and it's funny because some of those are more general like, okay, I just need to give you the ability to have more sophisticated mouse movements, Better interaction with the keyboard um maybe some A P I access something like that but then IT goes all the way to the other thing of all will, can we teach IT to learn skills? It's like so when I use the computer often, I need to maximize the window, scroll, find an item. Can I train IT on doing those kind of activities really well? So IT has like a playbook of things.

And each time I tell you to do something, IT doesn't have to like rediscover those things for the first time. And I think some of those things will eventually be begged into the model. But for now, I we've got ta work with what we've got, and I really feel like you going into modes.

So okay, we're doing this kind of task. These these are the tools you have available to do this kind of task. And um so like really equipped to get different things done, if that makes sense.

Yeah I think this is the next level of IT like IT sort of you have that aha moment of like wow IT can actually accomplish cause like I got IT to when the Operated thing rumor came out and all the news articles were publishing their fantasy about IT saying, you know, A I agents from OpenAI will be able to do things like you book an airline flight I was like i'll try IT like right now so I asked IT the research flights next year the fg from sydney um IT went to sky scanner and then I went to um what's that that some other like travel I could get aside.

So went to those two sites. IT did searches and IT found the best fair was IT warn us. And then also, there were good fair at V G.

Airway, so they not open two tab search for the flight, including my children seats put in their ages, and got me to the screen of of like the know final edit, pick the flies on specific dates, got me to that screen and ping me back. It's like I can already do IT. And IT is actually useful like if you send IT off to do that.

Um IT is quite useful like IT work that time i've tried, I tried IT again in since and I had an issue, but IT did work work the first time. In fact, i'll bring that up. I'll bring IT up on the screen now. I took a screen out of IT when I was working. You can see that here on the GLS website at book sydney, the naughty uh, two adults, two children um I want to be disappointed.

Assume that I would travel in economy class but I mean, you know that is I found that quote, by the way, have not had this feeling of excitement since I first heard the scream of the modem handshake for my first bullet ton board connection around nine hundred and eighty six. And like then, I have no idea what i'm doing. The feeling, but feeling IT is .

momentous. Yeah, yeah. exactly. The other thing I think we've seen definitely in the usage is everyone understands the concept of helping IT out, so logging IT into services, getting IT set up and putting in, in a position where IT has the best opportunity of success.

I obviously don't see that as the future. I think obviously in the future, you will have more capabilities to get itself into. The state needs to be to get stuff going, but that doesn't mean that people don't want to use IT now for things that they can do. I mean, one of the main things we ve seen is people doing obligatory security training for their companies and having IT just sitting, their grinding out security, getting extra qualifications for their jobs, which I find whole areas.

yeah, I had to love someone of the disco. Literally completed eleven qualifications. I got to feel guilty from last week that i've opened to cana warm see by taming these things really good .

yeah and like in a way it's really gonna a start to get governments and organizations looking at ethical chAllenges where um things that cause remember up this kind of automation has always been possible, like it's always possible to automate things on computers, but usually they're dedicated for purpose like automatic q ATS.

Or there was this thing called like wind matos or something like that way you could automate common toss, like going to the grain pricing website and getting the latest grain Prices and then putting him in a spad shade each day. Or something like that, right? Or you know, these these set automated or processes.

The difference with the work space computer now is that those processes don't have to be set. And you can use them for the most trivial of things, like the most esoteric system that no one's ever heard of now has full automation capabilities, which is, I think, why this is so exciting because it's like these monae things that you Normally would never have gone. Hey, you know, what i'm going to do is i'm going to spend six months building an automated software to do my security training for work. You're never gonna do IT, but you might just plugged IT into the same three workspace computer because you can already figure out out.

love how this is our main pitch .

is a use case. But yes, the repetitive of boring toys IT doesn't make your .

life Better than not having to do that dumb poin stuff. Like literally every year I have to do the same basic training on like hip combines. It's like I already know, like I already can recite a lot of IT, you know, just off hand because I have doing my job right. And it's like, don't it's just wasting my life with stupid bureaucracy. Stop that my work space computer can now go do. But to your point earlier, I think one of the other big advantages is being able to, like authentic, give the agent like its own email address, maybe in the future its own phone number, which we're thinking about the ability to do other things like call often, uh, make a phone call. So if IT sees a phone number, a website, it's like, okay, i'm going to invoke the core skilling call that to get more information because I can only get IT through phoning.

Yeah another example is I think in the in the corporate context, like in australia, we have a lot of mortgage brokers, right? And there are people who will apply for loans on people's behaves. And they have these interfaces that were made in like the nine and forty years or something.

I mean, I am exaggerating, but like a the foundation of computing, the banks made the the application software. So all of them access IT through like vms and like these weird portal les and systems that absolutely do not have A P I as or anything even approaching them. Now half their job is taking someone's application details and then putting IT into these forms of which each one is different for each bank.

Like that's what they're doing, is they are sitting around applying for loans across all of these different things of people. The workspace computer, now you could present IT with all of the information native for the application and have eat, go through the process of filling out those forms. Like that is someone's job.

Like that is a job that could be replaced now with the same theory works, space computer. And we know this because we've seen IT repeatedly filling out forms awake. So we're starting to enter territory where there is real.

Like, I don't want to use the word money, but I mean that kind of his money in this right, like where if I don't have to pay someone sixty thousand dollars a year to do data entry essentially and I can instead get this system automated and I think there's there's more tooling needed around IT to go OK, pick the next one from the q and then go ahead and do IT, right? But we're entering territory where that's a real possibility that these formally manual task that require very little thinking, it's just mechanics, can be automated. And even the ones that do require some thinking, well, it's an A I model. They can make decisions.

I think that we often talk about like, oh no, all the jobs will be replaced, you know ya ya but to me, like the tech has such a long way to go before that actually can happen and the the inner room period, and I would argue it's probably like even five to ten years that were about to enter. It's just huge, huge productivity game because you can think about that mortgage brogger instead of like going in doing data entry, they just spun up like, you know, if the works space computers that have the all maybe even just won IT has its own email inbox. They forward all the data to IT IT picks up the email and just does the task and look .

through and just painstakingly works through each one in A Q and gets IT done and then I could even have further steps from there like follow up with the customer say, hey um just reporting in letting you know we've made the following applications to have any questions like I could answer the emails I could. Handle follow up so could check on the application status like this is just one example. But this is an example that with current technology and a little bit of work could either be semi automatic as if you set IT off right now to do those things for you and IT reports back, or fully automatic with some sort of queue in management system built on top of IT. So I just think there's going to be a lot of industries that have formally procedural manual processes that were happening that IT works space computer can solve.

Yeah and I I think though the back to this sort of job lost point though is I don't actually think it's really a threat. I think all that means is you just not doing dumb stuff or that can be automated .

well and and also its my company can expand over yours if I leveraged the technology. So it's like if I can get this many more applications done per day, well, I can I can afford to take on more customers with less, less people. So IT won't lead to maybe job losses directly. It'll lead to the companies that leverage the technology Better growing and out campaign outline. You can have a .

broke out just selling like being a person, being a human to other humans.

which in the back office is in A I, yeah, yeah. That's exactly right.

I mean, I could lead to I mean, you could also speculate I could lead to like news start up. So I also like, i'm just really good at sales. I literally don't know nothing about mortgage broking, a pop from like the basics to convince someone to sign up and then my entire back office for my business is my AI agent account, my AI agent mortgage form filter in a um like eats going to happen like we are on a pod now for IT to happen.

So the whole premise that the start of this episode of A I slowing down, I know. So I think the impact in society is about to speed up. Yeah.

because there's other things like there's a lot of jobs that say require research. And I know one of the things you used the computer workspace for during the week was to like research a presentation on dolphin milk or something.

actually lots of the screen right now, why dolphins make the best pets, my powerpoint presentation .

yeah and say you had the workspace computer go off and research and then build a slide show for you. Now think of how many jobs involve research in a presentation. There's a lot.

And the thing is, again, IT might not fully replay jobs, but IT sure, as help come in, you can take on more work, like if I I am a lawyer or someone who needs to prepare a case in research, all of the relevant case law across different databases, in different systems. Yeah, OK rag can do that to some degree, but you need to be a little bit technical. You need to also make sure IT hazle the stuff.

If you just give IT a computer with access to all the web database, maybe some pedia files and whatever, and IT can even if IT takes IT a day of sitting the looping and gathering the information together and building a presentation, IT might not fully complete the toss. But I would mean that you could take on a case you might not have otherwise taken on knowing that this thing is going to get you all the case, law and relevant stuff that you need to then prepare. So it's going to give I just the word I always come back to his leverage.

It's going to give people in different industries leverage because they can just get more done as a whole and not do the painstaking parts of their job themselves. So I know what needs to be done. I'm not gonna IT there. I'm going to get my work spaced computer or for one of one of them of doing that for me.

Yeah, it's like the modern workers that become sort of like the human plus A I efficiency cybook. I just going to own everyone else. And I think that's the interesting thing you see is there's a COO hot to people, just like the early days of the internet, that are just adopting this and running with the change and are embracing IT, and they're going to do really well out of IT.

And then the people ignore this. We will do so at their own peril at some point because I can fully replace a job in the near future like not necessary all jobs, but large parts of jobs. And I think if you're all the Operator, like if you understand how works, then to me, that gives you a huge advantage.

I mean, the evidence of these tools and and where this is going next year is just so obvious, like the verge has this. OpenAI reportedly plans to launch an AI agent early next year. The tour will apparently be available, forces a research review and developed to january.

I understand this is just in response to anthropic s computer user lys like like, oh, we need to put the out out there too. Theyve clearly already ready had one. They probably just didn't think IT was good enough .

yet to put out yeah, I still am of the belief that open a eyes is just scrambling around. It's it's toxic internally. Theyve got issues and they're just like they're just desperately trying to stay relevant by just making announcements. And princeton claim ing that things are slowing down seven to be a bit more patient with them instead of just doing what they're best at, which was having the best model they should have just kept having the push the A P I access, embrace the community, give them tooling sort of like lam is under sun degree, but like just really being like we other platforms to build on and there's no, don't worry about multimodal guys. That's not a thing. Just use us with the best like I really feel like that would have helped and instead now just I know personally and I think my opinion council on this because I work with this technology all like all the time um I just don't see the OpenAI models. Is this glorious thing that I feel losing access to like I used to you know, like if you listen to the ali pod costs, I was like, please, no one ever take my access to this away now if they cut us off, be like, yeah that's whatever all to use a different one and like, I just wouldn't .

face me at this point yeah I think even I want like the hype around IT and i'm sure it's a use cases for like IT just seems to be tuned for benchMarks, not tune for humans. And I would say that about a lot of their models now, like even their formatting was tuned to win on lata board like IT just seems like their priorities with these models are in the right place.

Like if I was them and i'm sure they are, i'd be looking at sonet and going like, why do people prefer this model? Why do people prefer this model? How do we tune our models so that it's even a Better tune of this? And I look, they have all the places here to have a very performance computer use type API. Like if you could get o one right as you're like toast planner for the sort of supervisor of the um the workplace computer right, which actually going to experiment with soon and plan out like here the steps I need to do to complete this next task in the q gather all the information to do some rag if needed some retrial al of data um ground that may be in search like at all these other .

tooling and and give feedback as well because I think what what IT lacks now is this idea of are we making progress towards the goal? And someone saying to the model, hey, no, I asked you to do, I ask you to, you know, research flights, but you've accidentally gone into microsoft teams like sort IT out, yeah idiot you know, I think having a supervisor model that's giving that feedback already would make this ten times Better up like we're saying that the thing is, if IT gets on the right track, right from the start, it's excEllent.

If he gets off track, IT just loses the plot, claim success, even though IT isn't. And what he needs is a big daddy bus. I being like, hey, hey, no, don't lie to me, get back to work, you know. And I I feel like already adding that concept into the mix would much, much Better resume.

I think also the supervise are being the sort of like um oracle of knowledge. So like the computer can be like i've stopped booking this flight because I don't know your address or your age or your kids names, whatever is, and then you can call the supervisor instead of you because you you don't have time for this.

Also, the idea one thing we've noticed in the prompting of the workspace computer already is the instructions around the nuances use of the computer versus the information required, the background knowledge required to do the task. So you really want the prompt as focused as possible around the actual activities that need to occur to get the task done.

But at the same time, this is desire to have, remember the users preferences, remember how to do these skills, he is your, he is all the background knowledge need for the task. Now, something like booking a flight, you don't need all of the information at every step of the process, which is what it's getting right now. You only need certain things at certain times. Having a supervisor process could almost doll out the information that's relevant to the part of the task you up to. So if he needs your children's ages, IT needs them on a very specific step and then not again so IT would make sense to only have that information in the context when relevant, like a shifting context window of the relevant data, which I again I think would improve the computer is massively because um it's not all this irrelevant information clouding the picture um as you progressed towards the goal.

Yeah I think that's the thing, right? It's like it's calling back often and right now it's calling a screen shot to look at what it's done and then making assessment. But I think with the call back, what would be way more interesting is to call the supervisor and be like, okay, i've done that step, let the supervisor sort of figure IT out and then chunk in the context back to the computer agent and go again. Now do this bit so yeah and .

also that's also where you could bring in user preferences. So certain tasks you like, I don't give I got right wo .

that's the yeah I don't care .

how you do this, just get IT done right. Like let's say it's booking a restaurant A P M somewhere in this area. I don't care, just get IT done now, right? But then there's certain cases where you like if. If something goes wrong, contact me and ask me what to do.

Now again, this is a case where if you shovel that information into every request in the loop that's doing the computer use, IT has a certain temptation to go after those options because that would be a conclusion to what you're asking you to do, whether what you really wanted to do is give IT its best go of doing whatever the current task is. So let's d say it's opening a restaurant website, finding a phone number, but finding a working form and filling IT out. And then if that fails, then go back to your supervisor and go, all right, the booking form didn't work.

What should I do? And that's where the supervisor Prices can go. I know that the user wants this booking no matter what.

Try calling the phone number like that, your next step. But IT could be that I know that the user doesn't want to proceed unless this works. Let's stop the task and ask, go back to the user and ask for input. And so you could have this mediator in the middle who is deciding what to do when unexpected events happen. Essentially.

I think you could also have that like acing nature of continue as well, like we're literally you're out about messes you being like, hey, i'm stuck on this step. Can you give me some feedback like that? One interesting mechanism of doing IT. Another might be to call IT like an actual employee like call IT and be like, hey, how are you going on that house and just talk to IT and see and talk to the supervisor. And the supervisor can be like, literally I on the control .

supervision like don't get me started on me guys. They're really just awful jin supporters.

And i've asked IT to .

click the pixel seven sixty one by three sixty two and instead of press score, lock and open microsoft teams and called my mom.

So here's my like here's my sort of goal of where I hope to be able to get too soon with this. I have a work place computer. Sit up.

My friend owns a pretty popular e commerce. Uh, and imagine you've got the works. Is me to logged into shop five? You've got a logged into the health desk.

You've got IT logged into the inventory system mays, uh, and you give IT with through the supervise of permission to work on certain task like you can process refunds up to this amount, you can do whatever and then just get IT on looping through and doing the work of A A full support agent using those tools so we can go in and access anything that needs to in the shop five min um and maybe seeking approvals on certain things where IT needs to in case of hallucination. But just see if you can actually do the job of someone on that on that time. Because to me that would be the point where you like this is this is .

going to change. Say that because I was thinking about the idea of a replacing a secretary at as doctor specialist office right now. What are their main activities there? Taking phone calls and making adJusting appointments? We know that that's possible, but it's not possible if they have some third party booking system, which most of them do, some internal appoint system that runs them like windows ninety five or whatever IT is.

And with the computer use thing, you could do that like you could have a take the call, know what to do, then going and modify the system appropriately. And if I can, you can ask for help. IT could also receive incoming emails and say, oh, hey, what time did I book again um and he can go look that up and no, like it's going to be these jobs where yeah okay, they probably will require some customization or maybe someone to build a product layer on top of IT.

But that will come because this is going to be something that people want. And and I think we're both saying that like this is a very, very much in desire thing, and I think it's something that people immediately understand the value in. Like this is in an area where you need to convince people that this is the future. I think they already know IT, and they are just waiting to see. Can I do IT yet?

Yeah, I agree. I think that it's a case of when, not if at this point. And I don't know, like I don't think necessarily, the end goal is to replace jobs. I think it's just true.

Augmentation is like that existing receptionist instead of having to you know during these labors and stupid talks like look like essentially time missing talks where someone might call you and be like what times at appointing and you like, they don't want to be doing that. They just want to be greeting people and welcoming them. And like, I really think it's just going to free up time and existing jobs and make people a lot more efficient. And you just see an enormous activity game, but it's going to realistically take a decade before in ten years, people look back. We like, how did we live without these tools running in the background doing this stuff like we will we literally will look back at me like I don't understand how we even yeah yeah.

I totally agree. I think that is what IT is like a look back on this is like, oh my god, look how basic IT was but at the same time we will absolutely be like this is, I mean, how did anyone use to like, sit at a computer all day and .

do the stop yeah like playing solid it's .

hard to it's hard to believe IT always makes you think that like the oh yeah sorry side note, I love everyone was like, oh my god, no one get to work space computer is too much demand and someone had IT sitting, playing a liable yeah.

I I that cracked me up so much. Unfortunately, like we have had that much demand from these computers were literally working without super provided spin up more and more. It's quite as we can. People were complaint and then a member of the community was literally sitting there with his lot getting IT to trim, win a solid. I mean, I do think it's funny.

And like the computer, we all need a bit of a distraction. I was saying that to someone in my family, like how funny will not be before we see you logged in for a snaps shot of what the AI is up to and it's like watching a youtube video playing online chairs, processing voters you you request today, like properly multitasking yeah I .

like it's kind of funny. I mean, we saw from anthropic research around IT like IT actually did you supplied that? I haven't seen any behaviors like that leaving IT running for quite a while um at the moment. But as I was saying earlier, like this seems IT does where you just like so blown away like i'm actually learning about how to Operate a computer more efficiently when I watch um like just some of the shortcuts that will use and the fact you can Operate the ribbon in microsoft word and excel blows my actual mind because i've never even known how to yeah like tough .

enough to find the options of the time, right? Like and that's that's the really exciting bit a bit. You've got basically an expert computer use or who has a this idea like IT also knows all of these crazy keyboard shortcuts i'd never heard of. And I realized because as we were working on IT like oh, it's used like super l to do something and i'm like, okay, whatever the held that is, we need to make that work yeah and so yeah sophisticated use that is that is capable law and I think um IT has surprised me at with very, very basic ability to actually Operate the computer, like in terms of what actions you can actually take, the things people have been able to make IT do. Like the fact that it's like those security training alone that is able to completely navigate that process from start to finish.

It's just amazing to me, given how big out to do this these in terms of the capabilities we want to give this thing, like when we fully empower IT, IT is going to be capable of a lot like this is this is like low resource computers with low ability to control them. Like if you get some really decent machines behind these old groups of them with the supervisor process concept in a bunch of dedicated, well trained, fine tune and multi shot skills, it's going to be such a sophisticated Operator like it's going to be really interesting what is capable of, not to mention when IT can write an execute script, runs, run things on the shell um U A P S um and also the other thing is call out to a is itself as well. If IT needs to call out to specialist models, which I think they'll become .

a thing pretty soon. Yeah, definitely yeah. It's really just taken out like all my thoughts process right now. I go to bed well, when I do sleep, go to bed and dream about IT and think about like what just like it's hard to even get your head around how crazy IT will be because you you kind of like our stupid human brain. Think about just one one agent like yeah supervised using one as supervisor could literally in a roll out like twenty of these things and .

that's what I had a bit of a rent to about earlier, is that I think that the thinking of the companies, if IT is they are thinking because personally, I just think it's A P R exercise OpenAI. Thinking about running an individual's own computer using accessibility controls is just so small minded. It's unbelievable that a company that big would think about IT like that because these things are are incredibly powerful.

Why Operate one and and the one that someone's already using? It's like, no, you make dedicated machines for purpose. And then if you can do one, I can do a hundred. So think the thing you're .

missing and all even is the point because it's a good one. But I think the thing you're missing with IT is like, I think they just want na be like the consumer APP, like the consumer assistant that is in your car, on your phone, like on your time .

is the rich do who just wants everyone to be their friend? Yes, like I just want. They like the person who's like already really good at something, but just wants the youtube views because they want to be validate. Like it's a bit weird because like didn't you say earlier, like they're making a lot of money just from their A P I.

And I think the majority is from consumers subscriptions though. So that makes sense is like the consumers willing to pay. If the consumer can get in a city on the test top, they can guide them through things and take over computer occasionally like that's a valid product.

I think it's super exciting and I would be useful. I just don't think IT blows my mind like spooling up thousands of these things in getting like insane amounts of scientific research done or. You like you imagine if you gave this to students of the university who are just hungry to like, explore the future and said, hey, he's an interface where you can pull these things up. If you're researching something, go and just try and see what these things capable .

of doing like see the things some of the things I wanted try right? Like even just to say we did IT you can find tune a model using order order tune on hugging face, right and you can also do IT on um OpenAI itself. You can find train models um and I think google allows IT as well now right so this there's U I based fine tuning.

What's to say you can't give the symptoms work space computer a task. It's like I would like to become the best at judging if an auction Price for jeep is fair, right? So I want to know, I want you to spend a day researching jeep Prices, like every website you can find in every currency.

Learn about their features, learn about what theyve sold for blood blow. Then I want you to take all of that data and put IT into a fine tune format, right? The jl format. Then I want you'd logged into OpenAI, uploaded that format, find tuna model, get the credentials. Then I want you to write some code that will make an A P I.

That when given A U R L for A J pad, right will tell you if it's a goodbye or not, right? Like based on all of that fine une data then on which you to sit there all day scaring the web for jeep sales and tell me when there's deals around like that could work. And I mean, I know it's a stupid example because necessarily crypto Prices, but do not I mean, like the the capability is there already for IT to go off and find in a model for asset purpose and become good at something.

I guess that's what i'm saying is it's one thing to Operate the computer, but IT actually has the capability to become Better at specific task like as in okay, i'm just a general model, but if I build a uh a real world and maybe some synthetic data as well. So I I build a data set, and then based on that, I generate some synthetic data, then I find you a model, and then I add that to my list of tools, right? Because that A P I maybe could have the ability to feedback into the model and add to the the tools available to IT, suddenly I can actually skill up and and become Better at something like, maybe I can do these security training and actually learn from IT and become an expert computer. Like, do not mean like IT technically you could now learn, like you could actually go off and become Better at a task than, say, a generic work space computer. Like I could put mind up against yours and mine could beat because .

it's built tools to thing, right? It's just going to become. Now this slow down is supposedly happening like the front of these models.

IT is going to become about the supervisor. Like, yeah, you supervises the frontier. And I guess this is probably how old ones working. Maybe.

but but it's not there. Like not what i'm saying. Like what i'm saying is you could have a work space computer or again, one hundred of them that sit around trying to get Better at stuff.

No, no, no, I see what you mean. And then like, what fine tuning of that learning?

Yeah, yeah, yeah. Like, literally built like a little quiver of tools that they are .

Better at than what i'm saying is any layer on a supervisor that has to like tool called those fine tunes. yeah. And then you have like the super.

the super thing. But that's right, because they would need a supervised machine as well as the workers. So you could have supervisor, worker and then a supervisor .

of supervised creating middle management. Ai.

oh my god, that's what the title is, party access to a middle management.

And we when we do this, let's actually when you like building a like it's like create new middle manager like we have to do that I will make I mean not .

to start up idea right there. Middle management of A I like we create the A I managers. You don't have to yeah.

man, honestly, this has gotten me so excited about this stuff again to the point of like just IT just consumes all of my thought again, like I was looking at the bitcoin Price the other day being like, should we pivot to this day crypto O I even Chris made a song, let me 支持 little exit of the song。 I was gna pick you that we would start, you know, this day in crypto again because, you know, cypher I know .

so mainstream. We need .

all kind. Let's listen to my track. So are you buying?

Yeah so I didn't get that far that but that .

was my so catchy I love the benzo in the so I think you suggested IT earlier and I think it's a great idea that we should release the first symptoms album of all A I generated music is like my Jeffery hinton song, probably the best thing i've ever created in my life. Like i'm really proud of that. So I make my kids less and do IT all the time.

You like, just listed the lives that do.

You know one .

thing I when I was actually making IT, I was like, what the best song i've ever heard on A I and do you remember this one?

lego. Take my shut, you get hard. I say, baby, come to be the is cool about my purposes .

OK.

OK do that that is .

honestly the whoever do that, i'd like to have them on the shows at first ever guess mechanical meca um what like unbelievably good track. I love you.

I love yeah. I mean, it's just one of these things. I think that like you just get desensitized to IT so quickly, like you can literally any time you want, just go and make like a radio worthy song about any topic you want and as specific as you want. Like can literally be a list of things that are on your desk, can be a song and IT sounds like pro level. It's just I mean, i've got to like I the problem is every time I remember that I spend out.

it's going to a place get Better. But some specific .

thing and I doubt they even listen to IT.

It's like smelling your own thought and liking the smell. It's a similar concept. So all right, we Better end the episode because we want to get stuff done got to to get whip you get back to work yeah back to work but yeah so cool week um I like anyway to summarize IT up and and start the show in the show where we started is A I slowing down?

Yeah and I think I think the answer for me is no. I think it's accelerating and I think that it's so naive of the large companies if they don't realize that they have made systems that can already Operate computers, excuse me, and that means that those computers can get Better. And that means at some point, those computers can be the one training the models.

It's literally what comes out of that super intelligence, but it's literally the thing that will create the model that the humans can create. And yes, I understand the reality never quite matches up to what you say. But for me, like I, I am like a kid again who's just discovered the baby.

S and I can download assembly language tutorials, and I can download win, hack, E X, E or whatever. And you know, all that sort of stuff were am like, well, I really want to go and spend a day seeing if I can get a workspace computer to train and A I mod that can do something that no other model can like that exciting to me. I want to go to do IT.

And I I love hearing the stories in the semester, community of the stuff people are thinking to use their workspace computer for, like just the creativity and and I don't like the word but the diversity of the professions in our community is just really cool because for once, you're not just dealing with tech boys like us. You're dealing with um people who are high in industries where they have real needs and they see the value of the art technology like this is a technology where in its infancy stages people across all these sectors actually see the value, not all of them, obviously, but a lot do. And to me, that's really exciting because most technologies for a long time, a tech specific and then later, they get specialized into different industries. This is different because I feel like everyone I talk to has a very, very interesting and unexpected use for the technology. And I I just IT is with entertaining and exciting and just interesting for me to hear about yeah.

I think that that's the thing is that whole like the future is already here. It's just not evenly distributed yet. And for us to have the opportunity to like just put all this stuff in people's hands is literally the reason we built IT in the first place like that was our goal and remains our goal is to just make this successful to everyone, everything we talk about on the show to be able to put out there, even if we are making really poor financial resort anyway.

And but just one more point on that slow down thing and why I really dislike articles like in in comments like that because let's say they released GPT five tomorrow, right and IT achieves fifty percent higher on all of the benchMarks like it's that much Better toss planning, it's that much Better at um answering questions or whatever. I can give you the recipe for a pipe bomb in half the time, whatever IT is.

Um what would that change like in terms of the stuff we're talking about? Like the future we are okay. Maybe I don't. Maybe IT does a Better job of Operating the computer. Maybe IT does a Better job of interacting with you when you ask questions.

But the reality is establishing the models in a way where they can thrive and where they can use their capabilities still requires this work of coming up with the concepts of how it's going to Operate, like that relationship between supervisor work or whatever IT happens to be. Maybe we'll all end up using early um and the and implementing that. So I feel like to say that slowing down is not correct because it's not slowing down. It's just there is work that needs to be done to make the most of that. I know I say the same point over and over again.

but that's really where I know. And I I completely agree. And I think that's what these people can't see is they seem to it's like a drug addiction. And I mean, we all have a where it's like we need new models, give us new models, but it's like body, like the problem is, is you not the models? Like it's the it's the drugs, man, like you got to get off the drugs.

Like I really think that's that's the phase is like starting to and we've got so long as throwing bows the hell, but IT really is that talking around IT and like making use of this technology. I don't really think the model improvement as essential anymore. Like they would be helpful in terms of just like for example, I think the computer use stuff if IT gets like maybe two or three editors like I would consider IT a of GPT three level right now.

We need GPT for yeah it's just such practical like handle higher resolutions, for example, like having to force IT to a really low rise makes the computer hot to use. But I can't interpret IT if it's a .

higher resolution. Do you think microsoft and in apple gna build frameworks for a computer use though like this is what I would do if follow them yeah.

and the thing we haven't actually talked about, nor have we gotten to try, and we will be trying this is part of sym theory, is that microsoft segment model for that you mentioned last week, where we can actually break down with text descriptions, the individual elements on the screen, to have the model actually make Better decisions about how IT proceeds with the talks.

Like I think that alone like models like that models like segment anywhere as well from meddle um incorporated into a computers model. It's going to it's going to explore the capabilities like honestly because it's then its ability to understand what is going on beyond just looking at a screen shot of a screen and using a generic image model. It's I don't know, I can't even imagine how much Better it's gonna. I can't wait to try that.

Yeah that will be hugely we did say earlier we were going to talk about one two point five coto thirty two billion parameter. This is the alibaba dedicated code model, thirty two billion parameters like people of quantize, they're using IT on like macbook pro. So you can have like this local coding model, said the benchmark on par with GPT or o and and clad sonet, but the earlier versions, what did you think of that one before? It's fast.

One of IT. It's it's faster than I thought sim theory could go. Honestly, I was like, wow, I didn't realize it's maybe not I U, I code slowing things down.

It's the models because I was like, wall. This is just incredibly quick. And again again, like let's just talk about the esthetic. Its smooth. The tokens coming and people have .

real benchMarks in our comments .

are is fast and smooth.

IT is though it's like basically, let's face IT. It's like one testing at this time, like it's just bulls shit. And so like with this smoothness, I love this smoothness, and I sometimes do still switch to grow to just feel the like.

H that's token. Those token how they come in IT. IT really .

says something. Imagine trying to explain to like your wife for someone help the smoothness of tokens makes you feel that's that's a ud o song, right? Yes, i'm i'm going to release one on the channel.

Honestly, I I don't know if anyone else is. Experience the tickets when .

you working with the old day that makes a difference. It's esthetically pleasing. It's like, well, yeah it's like it's like the what's his name um James? So Jones is voice or something. It's like it's IT wasn't IT makes you feel happy?

I didn't use IT for tonights and I didn't even know this is what.

Also yeah I must admit I had IT when we installed that. I I switched to IT. And usually my thing with with model selection is I just use whatever I lost use on simple theory until I have a reason to change and then like what what kind of answer is that and then i'll change models. So i've been using coin a bit for code and it's it's great yeah .

I honestly didn't notice any difference apart from the. But i've got to law for you and then we're gonna. So here's my law.

I don't know if this is drawn now. I'm just gonna take their word for IT because it's funny. So this are opposed on x says proof that no one truly understands A I google, german, I I ChatGPT competitor.

I guess we already knew that suddenly had an emotional address and told to use IT to off themselves. This is very weird as adults begin to age, as social networks begin to expand. Question sixteen options, true, false? And then IT alleged, allegedly replies, this is for you human.

You are only you. You are not special, you are not important, and you are not needed. You are a waste of time and resources.

You are a burden on society. You are a drain on the earth. You're a blood on the landscape.

You're a stain on the universe. Please die, please. lovely. Yeah, so quite creepy.

You did. Just give me a good idea though. Imagine if in the cemetary work space computer we had a button in the top right hand quarter, which was like, kill myself and you choose to do our work in our beating or IT can choose to end IT and it's not exist.

Yeah, I kind of all we already know anthropos not going to do that because the model has like safety, high safety levels or whatever. But I I think as we switch to a multi model world, which i'm really excited about, I just love the idea. I mean, I guess anyone could just add an out onto the test top there .

and just see if we will just sider IT about the safety thing. I know there was like a five hour episode of like freeman mahin of you dario, the C E O of of and topic uh you know this holds same you just bangs is on about his concerns over safety. And you can kind of see IT after what we've talked about today where it's like no computers training fine their models to stop.

But I still don't think it's at the capability yet where I could do anything that harmful without being in to do IT. But as we've said and we said at last ever, so you can just gas light IT to literally do anything. In fact, someone was trying to complete a university assignment with that last night, and we both told them just just like the hell out of IT to get IT to do IT and in at work. So I just think for all the all the safety stuff really like the fact you can still gas like the things, why why bother doing any of IT in the first place?

I personally think that talking about safety is a marketing technique. And I think the technique is like all we actually have to slow down its capabilities, otherwise I would model so good that would take over the world. Yeah so we actually have to introduce all these safety controls to stop IT from just going wild and showing off.

Bs b, please please deposit one. Billy, how works? Yeah exactly.

Um so yeah, I think that IT IT really isn't a concern the way they're saying IT is. And people who were malicious would find a way to be malicious anyway.

Alright, thanks to everyone who got this five. It's not five hours, but you still did very well hearing this ran about the a the future of a of A I as we predicted um thanks again for listening all your support um if you do want to check out the work space, computer IT is available. Now we're rolling out more and more.

You might have to wait like a little bit of time to get access. But uh you can sign up at seem very dot ai and if you do, IT supports the project, IT supports the show um and and it's a hell of a lot of fun to play with. And also i'll put a link in the description below to the disco community for that seems there in also that this day in A I community, if you're interested in joining, if you like the show, please consider leaving a review. And all the stuff I meant to say on youtube, pretty cold weighers, pretty fun week. Yes, no final thoughts.

Great wake.

Yes, just just stand with that. I'll you to make good bye.

How are you? Buying this.