What’s Coming For ChatGPT in 2025

Sa. Sa. Sa. All right, so welcome everybody to another stream. This one's going to be more like hanging out over a cup of coffee and discussing the future of what this space might bring over the next year, rather than like a tutorial, step by step based stream. We did a few of those over the week. You might have seen them, maybe. Let me just start by welcoming everybody and then we can recap what we did this week and what's coming up on Monday. That's the one that I'm, I personally most excited about. But the point is, Today is the two year anniversary of ChatGPT. Yay. Can we get some sound effects? Wait, it's not like I have a person for sound effects. I gotta press a button here. Let's do this. Yay. ChatGPT is two years old today. Okay, so this 30th of November, 2022 was the day when this thing dropped. It's the day when I found out of the application I was aware of. Yeah, let's not talk about like AI in general. Let's talk about ChatGPT here. I just wanted to point out that I actually stumbled across, I think it was, I don't remember which version it was, but I stumbled across Dall E a few months earlier and that was mind blowing. Like that was a moment. It was at a garden party in New York and somebody just showed me early access to Dall E that they had and I was like, wow, that was the first time when I kind of like dove into the AI rabbit hole. And then in one of my tech communities, ChatGPT actually popped up on a day of release. I dove right in. I was creating videos on two channels already. I had to cover it right away. I consider myself quite tech savvy with all the different softwares that I kind of learned over the years. So I quickly picked up certain skills, created videos. People on the Internet seemed to enjoy that, added more of that. Fast forward two years and we are here with the channel. That is my personal story with ChatGPT, but I know everybody has an individual one. And that's the most interesting thing about all of this, that everybody here in chat, all of these people, all of these wonderful people have their very own story. How they stumbled upon this, how they're using it in their everyday lives, how they're abusing it in certain situations, how they're generating advantages or literally just exploring, having fun. We see people, all the way from people who are retired in their pensions playing around with this and creating little stories and little diaries. A Little visual coloring books for their grandchildren, all the way to knowledge workers using this to double their output, triple their output in some cases. There's really a million use cases. But what I want to do today is not focus on the news cases. We do that week by week. What I want to do today is look forward a little bit, not look, look backwards. We do that a lot of times with challenges and use cases and hey, this week's news, you know what happened, what's new. But today should be about directing our vision towards 2025 and what's coming up. So again, if you're looking for step by step tutorials, little prompting techniques, how to build some, how to build the GPT, you can check out yesterday's stream if you want to look at no code automations, you can check out Wednesday's streams. If you want to check out the challenge that we issued and all the community activity, you can check out Monday stream or the upcoming Monday stream. We have all of that today. It's going to be a discussion and in this discussion I hope to involve all of you here in chat. We have so many, we have so many familiar faces also so many new faces coming in here. Disrespectful. Trying to figure out if there's a good AI that is better at philosophy. I don't know, can anybody help the man out? I think Matthias is suggesting suggestion here with Opus is pretty good. If you look at some, some pet projects, it's some interesting projects like Terminal of Truth that uses two copies of Opus in the background and it seems to be pretty good, but obviously not at the level that you might require here. Disrespectful. Other than that, good to see some familiar faces. Toyo, Gerhard, Stavros, all of these people. Aaron, we have Matthias and Dominique moderating here in chat. Some event hosts from within our community. Dom runs all the creative and visual and midjourney related events. Matthias runs a lot of the no code events, currently teaching people how to build no code chatbots in a video series. And everybody else, wonderful to have you here. But there's one more person. I actually have a guest on today's stream and we're going to tackle some of these questions together. I'm going to put some questions out there, some questions like, hey, what do you expect of ChatGPT in 2025? Where do you expect some of these agentic coding frameworks to end up in, I don't know, 12 to 24 months, what do you think? Or like. And I'm going to answer Them myself, our guest is hopefully going to answer them, our chat is going to answer them and we're going to have a nice little discussion just like a group of friends would have over a cup of coffee or maybe you know, a glass of wine on Saturday nights. Because that's what it is for me right now, Saturday night. And we're going to have a little discussion around all this. Okay? So with all that being said, I want to open things up with one question and then we're going to bring in our guest here and we're going to start our little discussion. So my question to you today is, you know, I always like to open up the stream with a question to the entire chat and today we're going to do. When did you discover ChatGPT? Because I think, you know, and there's no shame in also like discovering it later on, you know, people are busy, like it might just say that hey, you have more of a life than I do or than I did at that point in time. You know, I was like hyper obsessed with all technological things and I was just like on a route to trying to like self improve on as many dimensions as I can. One of them was learning software. So naturally I was plugged into many software folks communities and naturally I was, you know, I was sitting inside of Discord chats every single day like learning what I could. That's where I stumbled upon it. That's why I discovered another day of release. You know, if I had maybe more of a full life and some other dimensions, I wouldn't. So for me it was November 30, 2022 when I discovered it, when I heard about it for the first time. When did you guys hear about it? It. Which month? Which year? Please share it in chat, I would love to see it. Many people discovered it in 2024, you know, hopped on the bandwagon now trying to inform themselves, trying to improve their skills with it. But also many people here, I know, I know from the team, most people discovered it in early 23, kind of like spring 23. Matias Dom, like when did you guys, when did you guys stumble upon it? Matthias, I believe for you it would be like January 23rd I think. And then Dominique, I'm not exactly sure. I think similarly maybe, maybe spring 23. Yeah, so we have Wilt Alexander sharing November 23rd. Yeah, we have December 22nd HV yeah, December is December. I think if you discovered it in December, it says a lot about you. It says a lot about you and your preferences. So yeah, please Keep sharing that. And meanwhile, let me bring on our today's guest, Casey. I will put you on here. All right, so are you ready to go? I'll switch the scene. And here is today's guest, also referred to, Casey, also referred to as blazing Zebra on YouTube. What's up, man? How are you doing?

I'm great. How are you doing?

Oh, fantastic. As you can tell, like, we're having a great time with chat here, hanging out a little bit. And yeah, basically Casey is one of the YouTubers from our YouTube AI community. We have a nice little Discord group, very cozy atmosphere in here. And when I brought up the question of, hey, who wants to discuss what's next for all of these tools? Casey was one of the first people to jump up and say, hey, let's do this. So I was like, all right, looking forward to it. Yeah. Casey, want to tell us a little bit about, like, what's, you know, what kind of like just content you create and what your focus is or like what your vision is for your videos in general?

Yeah, I pretty much do how to. So just learning. I'll take something and try to dissect it and try to learn from it. Some of my. A lot of my tutorials are about how to use these tools to learn. You know, it started out in marketing because I ran a marketing agency for 10 years. It was a team of about 15, 20 people. And like you were saying when, when ChatGPT rolled out, it was December 2022, when I. It was about a week after it rolled out that my buddy texted me about it and I was just all over it and, you know, didn't really know what to do because my, My, my marketing agency, we focused on creating long forms, blog posts for SaaS companies. So this hit us. I kind of knew something like this was coming around. There were other tools that were getting better and better. I was watching GPT2 quite a bit. You know, there was a big New Yorker article about that. But then I, I couldn't look away. So I quickly rolled up my marketing agency, shut it down, and really started to focus on this as much as possible. I had been doing stuff in, in AI earlier just as a hobby, training, neural nets and different things. But yeah, so I started the channel and a lot of it was, you know, content created. But now it's getting more into, you know, automations and, and learning and just sort of anything that. That piques my interest, I guess.

Yeah, man. Yeah. So thank you so much for sharing. And that's also like, right Away Like I like that about your channel that you also focus more on the, and less on the. What is the, you know, the biggest news story or what is the biggest drama going on right now? I think there's value to that too. You know, there's these, there's a lot of these. I think a lot of channels focus on something that I like to call AI gossip. And then there's channels I think that are maybe a bit more like us, where we're like, okay, what are the how to's, what are the step by steps, what are the workflows, what are the automations? And you know, there's value to both of them. But it's good to have you on here as we do. We do similar things in that regard.

Exactly. Yeah. Yours was one of the first channels I came across. Yours and Dave Shapiro. I saw people talking about Dave in the, in the chat. I worked closely with him for many, many months. He was sort of my mentor for a while. Know him really well, so it's cool to see people mentioning him.

Amazing, amazing. Did he invite you into our little community?

It was actually Matt Wolf. I met up with Matt Wolf at a conference and, and he and I connected and that's, that's kind, that's how that connection happened.

Yeah, that's wonderful. Yeah. Actually Matt, I, I believe Matt and I were the two who started the Discord server. I think it was like, it was like in January 23rd or something. We're like, hey, we got, we gotta, we gotta have a place where we bring all of these like different creators together. And yeah, so it's been a, it's been a pretty, pretty wild ride. Matt was also one of the first people. Yeah, he was one of the people from the first cohort on YouTube, like really leading into this. Oh yeah, yeah, yeah, yeah. So amazing, so amazing to have you on here. Also interesting to hear, hear about your early experiences with it. And like what personally resonated with me there was you pointing out that you kind of immediately rolled up your agency. Like, what came to my mind is that I don't think I was like very conscious of it at the time, but I remember when I saw the Dall E demo instantly it occurred to me that, hey, probably doing a video production company over the next years is. It's not very future proof, let's just put it that way. And also offering social media marketing to companies and everything. I was like, hey, I gotta move up the stack. That's what I thought to myself. I can't just Be producing for others because all of a sudden some of these tools will be able to. To do that and automate that really well. Clearly from like, hey, like, like these tools are insane. And if I. And if I manage to get in front of the camera and you know, like, like do. Do all that which comes with its own problems, then I will maybe move a few years ahead of the trend. Like, who knows? Like already we're creating all these avatars. Like anybody on this channel, I think everybody in chat has seen the little quiz that we prepared. We prepared a little quiz where you have like eight different outcomes. And at the end of the quiz you get like AI resources that we have free resources and there's an AI avatar of me presenting to you why you should use them and like how to use them and stuff. And man, you know yourself, like, these tools are getting really good. The voices are indistinguishable. The avatars are like a six or seven out of ten. Right? Yeah. So we might be already on our way out, but until then, but until then, we're going to be live streaming. We're going to be creating content that we're going to be discussing.

Exactly right.

We're going to be discussing anybody who.

Knows how to do that already and, and you know, clone themselves and create 10 channels. Be ahead of the game, I guess. I don't know. I don't think anybody knows. It's like everybody I talk to is like, you know, you talk to a coder. You know, one of my friends is way up into building some of these massive models and he was like, my job isn't safe. And then you talk to, you know, anybody from that to like a social media, a person to an attorney. And it's like every. Everybody's feeling that, but I think just the more you use it, the more you're ready and you start to see things. And maybe this gets into our 2025 predictions. They're starting to. I feel like there's things where these, these, you know, it's going to take other breakthroughs to really be this, you know, the large language model, obviously very, very powerful. But in my, my opinion is, you know, to get to some of these where like, you know, some of the really futuristic things that people talk about I think is gonna. Not many breakthroughs. Maybe it's just one or two.

Yeah.

And that could happen next year. I don't know when it's gonna happen. But you know, there's certain things that large language models don't do especially well. That if you spend a long time with them, it starts to become apparent, I think.

Absolutely. So like, you know, you just don't have the 100 reliability that, that you kind of need in many use cases. I think that's one of the big ones. Ones.

Right. Even if it's 95%, that's still wrong 1 out of 20 times and it's like not a good experience.

Yeah, yeah, yeah, yeah. And like wrong in a very confident way. So, so. All right, so, so you brought up some good points there. Like one of them is this like, you know, breakthrough needed versus just something we already have improving is I think the other bucket. So maybe we can start our discussion and, and you also brought up these like agentic coding frameworks and how they have been developing. So maybe that's where we can start our discussion. Again. Just for anybody who just came in, for all the new viewers, we're basically doing a little discussion on what's next for these tools. What's next for ChatGPT? What's next for these agentic coding frameworks? What's next for the creative tools that, you know, largely seem to be struggling to find a concrete use case? I'm concretely looking at all the video generators, you know, like, like they're really impressive but like what do you really use them for? Kanye west makes music videos with them. But like beyond that, like what are they going to evolve into? I have some thoughts here and these are all the things we're going to be discussing today. Also welcome to the channel Mahesh. Great to have you as a member here. Yeah, happy to go into some of the questions here, but first let me just kick off the discussion by maybe diving into these agentic coding frameworks. As you seem to be interested that you have a background in this and I know you have some thoughts on this. I do too. We actually inside of our membership we had a founder from one of these big, big agentic coding frameworks come in and we had a one hour long discussion where basically my main line of questioning to him was around where do you see this going next? Like how is this going to affect both coders but also companies that need all these functionalities. And basically I can tell you like, like the executive summary is that just the value of software is going to be cut into like, like by, by a factor of 1,000 or 10,000. And a lot of things that currently you need to pay 20, 30, $40 per seat for are just going. You're going to be able to hire a freelancer that is like whatever dev like in his case, like the company was Pythagora. They also like sponsored as at one point. As at one point. I'm sorry. So I just want to point that out. But there's also Replit Agent, there's also Bolt, there's also, I think, I think there's also obviously Cursor and there's Windsurf from Codium. So there's many of these popping up and they're super useful already. But obviously this is, this was kind of just the starting shots to this race. And if you really play this out, then you will have one person who will be able to build custom tools for companies, for corporations, and if they need a feature, they pay him $200, $500 and they. He just adds the feature and they own the whole thing. They don't need to pay subscriptions. There's going to be a whole new economy of people. It's going to be sort of like a.

And it'll probably work better for them than, you know, not only the, not only the cost, but the use case, you know, the implementation will be completely customized.

Exactly. So let me just, you know, that was a lot of, that was a lot of points there that I brought up. But let me just ask you this. Like, what do you think is something like a Bolt or Replit Agent or any one of these, like a cursor going to look like if you give it another 12 months of testing of them, gathering data of them, developing new features without any major breakthroughs, let's not assume that there's going to be some, you know, seismic shift in what's possible. But what do you think are these going to develop into? What are they going to look like a year from now? Do you have any ideas?

Well, I think this was right in my notes that I had that. I think we're at the very tip of the iceberg for, for this and that these tools are just going to get better and better. And the skill that I'm actively working on developing would be sort of the pro product manager skill. So instead of learning the nuances of Python, which I love, I have a good time learning it, but I sort of feel like my time is better spent these days learning how to plan out, you know, how a piece of software should work and how all of the different, that, you know, objects should, should relate to each other and that the AI will be able to take care of the coding fairly quickly. But I feel like there's going to be a. You know, I worked with many SaaS companies. I worked a lot with Salesforce and had a couple unicorns that we helped get from early stage startup to, you know, unicorn status. So I've been around the SaaS world quite a lot with my agency and I feel like they're gonna go through a transition that, that like major media has gone through in the past, whatever decade or two decades where, you know, when blogs first came out, when you first heard that word blog, it was like, oh, it's this funny little thing where I, you know, journal about my day or something like that. And then there becomes like Huffington Post or there becomes these ones where it's like, oh, wait a minute, it's no longer just, just like this little hobby. So I think we're like way on that early side of this. There's a term I like to, that I've heard, I don't know where it came from, but micro SaaS. So instead of this Salesforce, you're going to be able to, you know, anybody who is good with spreadsheets, you know, before will be able to spin up their own sales force, you know, for their, you know, medium or even large size company a lot faster than they could even implement a paid version of Salesforce. So I feel like that's the model as, as we've watched mainstream media now is no longer really that valid. Too many people. I think the same thing is going to happen with these big SaaS companies with, you know, you know, for sure, like Salesforce, where it's just sort of a database but even all the way to like zoom or different, you know, discords and these types of things could, you know, could be disrupted and then there's a ton of opportunity for like the AI hacker, the person that knows a little bit of code, a lot about project management. I think that could really open up and be a massive, really fulfilling career.

Oh yeah, yeah. So I think this is, I fully agree with that. I think there's a space is going to open up between what currently are like. I think Slack is a good example. You know, it's just a communications app. App, it's just a chat app. Nevertheless, if you have hundreds of employees, you have a $10,000 per month bill just to have access to Slack. Like there's going to be a point, I think in the not too far future I would even, you know, make the assumption that this might be realistic within a year or so where you can hire some like, like the founder referred to it as it's going to be. It's going to be Like a new coding language, you know, so now you have a Python dev, for example, but then you're going to have a Bolt dev, so to say, and the Bolt dev is going to offer his services all across the Internet or maybe even has some, have some, a little bit of track record and you're going to be able to pay him 2, $3,000 to build you a simple version of Slack that has all the security measures. Because you know, that's one of the things I think they're going to not just like, I know for a fact that they're already looking at the security as one of the core things to be solved here. Because, because all these apps right now don't really, they're not waterproof. Let's just say that in terms of security. But like within a year I think that's going to be figured out. So instead of paying $10,000 a month to Slack as a, like a medium sized company, you're going to be able to hire a dev for a few thousand and he's going to build a custom Slack for you from some sort of like SaaS boilerplates that already exists inside of Bold and he's going to add a few features and he's going to hook it up to everything in your company. Everybody's going to have it. And there you go, right there you have a saving in the six digits per year for a lot of companies. That's just a very simple example. But my point is there's just certain building blocks, blocks that are not in place yet. Why this is not possible yet. Like the security aspect, like a lot of the, a lot of the security surrounding these quickly built apps, they're just not on a level where you could trust it with, with a few hundred employees. Whereas Salesforce does provide that and you can just have that peace of mind. But I think within a year it's going to be there and you're going to have a new class of developer that might not even know how to code properly. Let's be real. Maybe you, maybe he just knows the basics, he knows how these different concepts work in theory. And he knows how to use a tool like Bolt or Replit Agent or whatever really well. And then he can build something that might cost 100x more. So that would be my. Well, that's not even a prediction. It just seems so clear that this is happening unfolding right in front of our eyes right now.

Yeah, and I think there still will be a learning curve where it won't necessarily be the code and I think it'll be more accessible, but there will be, you know, there's a lot of different hacks out there, you know. AI Jason. Do you ever watch his channel?

Oh yeah, yeah, he's great. Super unique stuff, huh?

Yeah, he released one of the cursor workflow that. I highly recommend that video because he talks about, you know, coding. I always would mess with it and start like, hey, let's write a script. You know, it's the first thing. Let's like write a program that does X and then you start perfecting it and doing better and better with it. He kind of takes the reverse approach where he uses the AI to plan everything out and to go so far as to build out the, you know, the folder structure before he writes even any code at all. And then kind of fills in, you know, it's like, it's like a coloring book almost where you draw the thing first and then you fill in all of the, you know, the code, the different things that, the scripts that would go into each folder. And so that's a skill that I think folks are going to need to develop. That's very different than coding was. So there's still a lot to learn and there's no one to teach you right now. A lot of people ask me like, what books should I get?

And it's like, it's like I don't yet.

Exactly.

Yeah, yeah, yeah. Very, very. I agree. Very interesting. So a lot of that is kind of like having a little bit of vision rather than that then like concrete, concrete skills, like software skills.

Creativity. Yeah. Innovation, seeing what's possible. How, you know, so many people I talk to like, oh, I wish it could just, just do this. It's like, well, wait a minute. Focus on all the stuff that it can do, you know, because you're always going to want some more and don't get caught in like as soon as it can, you know, be truly agentic, then my business model will be, you know, validated or something. But it's like we would have killed for this type of thing just whatever two years ago. Right. So focus on what, what it's good at now and that other stuff will come sooner or later and you'll be ready for it.

Yeah, it's true. Just the problem with that is like often people, they see something come out and so much of it is very heavy on the marketing, very heavy handed on the marketing that there's a disappointment that happens once they get their hands on it. And then, and then you kind of like, you know, I think, I think just not to, not to dive into this topic, but I think the crypto space is a great example for that where it's been just like, over promise, over promise, over promise. And then even if somebody like dove into that rabbit hole and got their hands on some of the protocols and was like, oh, let me transfer eth from one wallet to another and then you get hit over the head with like a $80 fee while transferring $50. And you're like, hold up, this is supposed to be the future of money. Like, what are you talking about? I just, I sent $50 and it cost me 80. This is ridiculous. And then like, then you tap out and you form an opinion on the thing. I think a lot of that might be a bit more extreme, but I think a lot of that happened with for example, advanced voice mode with ChatGPT this year because it was like, whoa, like it's gonna, you know, we're gonna have a universal translator that can use your camera. It will help blind people see and navigate the world, no problem. And then, you know, radio silence. And when it came out, it basically was lacking 80% of the features that they were advertising. Like all of the insane stuff was missing. And what was, there was a better voice model that, you know, could imitate certain voices well. But then it was pretty lobotomized too. So it's like mismatched expectations and promises.

Yeah, so much of it, you know, the expectations are so high and you get in there and you know, so many of these tools. I was just thinking about the rabbit, the R1. Remember that?

Oh yeah, yeah, yeah.

Everybody was for that and I was like, I don't know if that's really what they're showing.

Yeah.

And I don't know. So I hope the industry learns from that because it's, I think, you know, it's, it for sure in the long run is going to hurt the industry if they keep overhyping and under delivering, giving out these half baked things and you know, rushing to, to market. But the, the bitcoin or the, the crypto analogy is a good one. It's one I thought a lot about even just thinking about this call because you know, you could, you could sort of draw a line between bitcoin and chat GPT and then there's all these other things which are like the, you know, the altcoins or whatever and.

Oh yeah, yeah, I never thought of that. Yeah, yeah, you're right.

And it's like, well, are they really going to be able to do that? Thing that they talk about. They kind of can. On paper they certainly can. But like, is anybody going to use it? You know, Ethereum is supposed to be this like world computer. But did that ever really happen? I don't know.

Definitely not. In a way people portrayed it. I think that's a great analogy. Yeah, you kind of have ChatGPT is the Bitcoin and then you have all of these layer ones like Claude and Gemini and Llama and whatever, sort of, okay, they're legitimate and they work. And then you have all this coming with all the different things that pop up. The difference being here that obviously there is some sort of value behind some of the subscriptions. Like, hey, no doubt they figured out a lot of the interfaces and stuff. But yeah, it's just the marketing can be a problem. Okay, so let's get back to the practicalities of it. So we talked a little bit about these agentic frameworks. I want to point out one more thing that was really interesting that I learned in my interview with one of the founders of these companies. And he pointed out that like recently actually the release of Bolt has been, in everybody's eyes, has been a real breakthrough in terms of what's possible with front ends of applications because it just works fantastically well for front end only applications. And that they actually shifted their entire thinking about how to develop the next version of their tool where it's going to be two separate processes, one for the front end, where they do something similar into Bolt and one completely separate set of agents that build the backend and then it, then they have to make it, plug it into, make it, make it work together. And that's the big challenge. So that's another. So like just right there, there's a clear roadmap to making these tools way better than they are right now. And they're already quite capable. So I just kind of like wanted to highlight that next to everything else we've been talking about, talking about, these.

Things are going to go really good. What have you thought much about interpretability? So like Claude I think is leading and Anthropic is leading this charge where they're trying to figure out what the heck is going on in this black box of, of the large language model. And that I think is probably going to be more important in 2025 than the large language models getting better and better. You know, they, they will, but it's like, you know, what does that really get us sometimes, yeah, it's marginal or even if it's way better. You know, people are struggling to put, you know, the O1 preview to good use sometimes. So if it gets 10x better than that, you know, obviously it'll be useful. But I think understanding what's going on behind the scenes, if Anthropic can crack that, I'm guessing the other labs are working on it as well. But Anthropic is, I think has been the most public about it because it goes with their safety, you know, their safety, whatever focus and you know, understanding. And I think they are making some progress into what's going on in there and then in that way they can control it better, you know, so you can get to some of those more reliable outcomes, you know, that'll be repeatable. You know that you can set up some great custom instructions and have a good output every time, not just, you know, three out of five times or something like that. So that I'm really excited about that. If I'm hopeful that that interpretability, you know, if that gets 2x or 10x better than it is now, that could be some major progress in, in how we control these and the whole, all the safety issues and things like that, alignment issues, go, go, go along with that as well. So.

Okay, that's, that's an interesting point that I didn't even have on my, my list of like talking points today. Yeah, they're the ones that are very public about this. It's weird. Like there just have to laugh because like their marketing campaigns are like, hey, if, you know, you know, and then a Claude logo and like. Okay, okay, what does this like their marketing is quite terrible to the outside world, I would say. But then like on the, on the nerdy stuff and the stuff that, that we like to talk about here, they're, they're leading in, in many ways. So I don't know, I, I think you summed it up pretty well there. And I have, I have one more thing that I would like to point out and that relates to Claude. Have you gotten a chance or like, have you heard about the model context protocol that they released a few days ago? This is something completely new.

I've heard of it, but I don't, I haven't, I don't know much about it.

Okay, so we are just getting into this with the team and I've also, my knowledge maybe constitutes of like 30 minutes of poking around in this thing, but basically it's a way to hook up external tooling to a LLM interface. So you can think of it as like just using Claude or using ChatGPT and then underneath you have a little dropdown of different tools that you can directly connect to it. Like an external action that works directly with your chat. And so like we already studied like some research tasks within the AI advantage to like try this thing out and throw different things at it. But I think open sourcing something like this is pretty amazing. Just like open sourcing computer use is pretty amazing. Have you gotten a chance to look at computer use or like, have you?

I got in there a little bit and realized that like what it was really good at wasn't really a problem I was facing currently.

You know, it's not, it's not good at anything. Yeah, I mean, let me spoil it for you.

Yeah, I was like, I don't really need to like organize my files on my computer. You know, it's like that was sort of like the best case scenario. So, yeah, those are the things where there's, there's so much I think we could just do with the base models. It's like I could focus for a year even if nothing got any better on what we have currently.

Yeah.

So I, I tend to wait a little bit. Maybe it's a mistake, but I, I'll wait, you know, a few weeks until I see something really getting traction. Then I'll dive in and be like, okay, that's really cool, you know, because like you said, that hype, that hype just has gotten me too many times. And I spend, you know, three days getting into something and I'm realizing this really isn't ready for prime time yet.

One dude at this computer, you sing. It was a bit of a disappointment because I was really like, Claude didn't hype it up, but I hyped it up to myself. Like when I saw it, I was like, this is incredible. This is the type of thing I've been waiting for from OpenAI. And then we got in there, I even scheduled the event for it in the community. I was like, hey, we're going to do a deep dive on computer use. We're going to find all the things this thing can do. And then it was so sobering because the result was basically even the most basic use cases that we found for this. Like, hey, transferring data from a page page to Excel sheet or filling out a form online. It's so damn unreliable that I would deem it useless. It's super interesting, it's cool to see, but it couldn't even like, like one of the few use cases that we tried was like, hey, just go on these like different NBA sites. And transfer the data into a sheet, into a. Into a spreadsheet. And I ran it like 15 times. And five out of those 15 times, it just. It just spawned up random stuff. And it started doing, like. It mislabeled the columns and, like, started a second sheet instead of doing the first one. And I was like, okay, I need better prompting. So I worked on the prompting, worked a bit better, but then it went off the rails in completely different ways. I was like, okay, maybe I need to flesh out the process step by step. I did that with the better prompting. And then it started, like, hallucinating completely new things. I was like, okay, I'm giving up. This is not worth 100 hours of my time. It does not work yet. And let's. Let's hope it works better in the future. Yeah. So just quick share on my experience with that thing.

Yeah, that's. It's similar to mine. Like, my buddy and I, we were texting each other. He's building this software tool, and that's the. Exactly. You know, I think everybody's waiting for that. It's almost that promise of that true full 100 agentic. You know, if you can control your computer, you can get it to do. Do a lot of things that it can't do now. So my buddy was. We were. We were actually texting each other, like, at the same time when it first came out.

Yeah.

And. And then I went for a walk and I loaded in, you know, the announcement, and I. I kind of just was. I love to just chat with these different tools, whether it's, you know, I pretty much use chat GPT while. While I'm walking to explore. I kept asking it, what is this good at? What's the best case scenario? And it kept coming back with like, well, it can kind of sort of, you know, move some files around on your computer maybe. And I was like, okay, I'm gonna wait for a better. A better selling point before I dive in.

Oh, man. It's really. Yeah. I think it hits the vision that a lot of us here in chat and in this room are exactly like. That's exactly what we're looking for. We're looking for the. This agentic future where, you know, you come up with a task. You're the mastermind, sitting there with your little cat, and you're like, hey, agents, go do this. And they just figure out a way to do it and they get it done. But that day is not today, that's for sure.

I know. I was just setting up a mailchimp account. And I was like, I really just wish it could do this. It doesn't need to do anything more than, like, you just set up a mailchimp for me and then maybe connect it to Patreon. You know, it's like, not that hard, but we're still a ways away from that. And that's where I feel like we almost will need another breakthrough. I don't know that large language models alone are going to really be able to, like, do some of this stuff. Maybe people disagree with me, but it's like we've been waiting for this agent now for about a year and a half. I feel like everyone's like, it's right around the corner. It's right around the corner. And don't get me wrong, it can do really cool stuff, especially if you, you know, connect it with, make or, or zappy or get fancy with that stuff, but just like on its own, in the wild, autonomously. And then once it can do that, like, is that AGI? And then once it's AGI, how soon is it going to be super intelligent? Probably pretty quickly after it.

So, you know, interesting point there. Do you feel like we moved forward in 2024? Do you feel like we itched closer towards that vision? Or do you think that kind of like GPT4? Because a lot of people are making the argument that GPT4 was even more capable than GPT4. Always now, you know, so, like the last 12, sure, like, we have more token output and we lowered the prices. But, like, just in terms of, like, ChatGPT output quality, it might not be the biggest change. What is? That's just like, playing devil's advocate here, you know.

But, like, what's your take? I, I, I would say, you know, I tend to believe that they've gotten better. It's just, but it's in our, it's like, but, but it's, you know, marginal. It hasn't been that, like, you remember that graphic where it's like, here's a tiny pixel and then this is like.

A. Oh, yeah, GPT4. Like this.

Yeah. And then it didn't feel, it didn't feel quite like that to me. The biggest wins, I think, are like, the ux. So, like, you know, I think artifacts and quad projects was the first, like, UX thing where I was like, oh, this is great.

Yeah.

And then, you know, there's the CANVA feature. And I would say that, like, Notebook LM is basically just a different UX for, for a large language model and that type of stuff. I think that's where I'm gonna, if I had to place a bet, I think next year is going to be an explosion of all these different ux. Because, you know, I thought artifacts was great. I was like, how could this get any better? And then of course, you know, the canva came out and it's like, oh, now you can edit the artifact. I didn't even think of that. Or you could like highlight a piece and focus all of its compute just on that paragraph. Say, like, fix this paragraph or this snippet of code or whatever. So that type of thing, not only from the large labs, but from these, you know, AI hackers, from people who are in there building, I think, you know, the, the wrapper app of just like some good instructions and things like that, that that's going to go away and be replaced by these smarter, you know, different UX applications.

Yeah. So, so here's my thing that I, you know, it's just a fact, but I say this like all the time. Basically all of the hackers are gonna figure out the, the best, like niche use cases and the ones that like really succeed. OpenAI is just gonna, you know, they're the whale that just opens up the mouth and they're gonna quick, slowly move ahead and just.