ChatGPT Sol vs Claude Code in Club Jam

Tonight is the main event, SaaL versus Fable-5, the App Idea Challenge. So we're not just gonna talk about AI, we're gonna put it to the test live. And that means goodness knows what could happen. We got lucky last time when we did Claude Code versus Manus, but then I did Claude Code versus Manus for another community a week later and all hell broke loose. So who knows what will happen tonight? So yeah, we're gonna give both AIs the exact same prompt and watch them build. But first we ask them both to come up with the ideas. So I gave Claude and OpenAI the context and asked them both the same task: suggest 6 apps we could build live in under an hour that would be interesting, impressive, and give the audience something to vote on. Okay. And yeah, what was came back was definitely interesting because they have very different ways of working. So what Claude suggested was Roast My Idea, the business critic, paste a business idea, get a punchy critique with strengths, risks, and a score. Then a live poll voting app, audience scans a QR code, votes in real time, results animate on screen. Then a personal pricing calculator. You input cost, time, and desired margin and get a recommended price with breakdown. And then AI meeting note summarizer. There's already so many of those that already exist. Oh, let's see what's happening in the chat. Okay, the chat's a little bit too far away for me to see.

Nothing yet. Just some chat.

Okay, great.

Thanks.

Yeah.

Yeah. It's over there on that monitor. Build my elevator pitch. Answer 4 questions, get 3 pitch variants, formal, casual, and funny. And then the Habit Street Tracker. Simple app with a satisfying visual street counter. I can never remember what CRUD is. I think it's read, write, delete, or something, or copy. Justin probably knows. Okay. What was it? Sorry. Oh, CRUD. Can you remember what that stands for?

Put you on the spot.

I do not know.

Sorry. Oh, okay. I know every time I look at it and look it up and then I forget it. It's just some basic things that We do. I don't know. I know D is delete.

Create, read, update, delete.

Great. That's right. That's probably why I can never remember it.

Okay.

So now Sol came up with what I thought was some pretty interesting out-of-the-box ideas. So number one was Crisis City. A miniature city management game where the player responds to cascading emergencies: fires, blackouts, flooding, traffic, and hospital overload. And it should have an interactive city map with animated incidents, limited budget and emergency units. Decisions affect safety, public trust, and infrastructure. And you have a 3-minute scenario with a final score. And then with Conspiracy Board, which is an interactive detective board for solving a fictional mystery. It has evidence cards, suspects, locations, and timelines. You drag cards onto a corkboard and connect them with string. You unlock clues by making correct connections, and you present your theory finale that evaluates the accusation. Then there's Beet Garden, which is a musical sequence where every instrument appears as a growing animated plant. You click a grid to plant drum, bass, melody, and effect seeds, plants pulse when their notes play, tempo, scale, and mood controls, save or randomize a composition. These are things that I've never seen in any other thing like ChatGPT or Gemini or Claude. They're kind of very out of the box. Evolution Lab, a playful ecosystem simulator populated by procedurally generated creatures. Creatures Speed, size, vision, energy, and camouflage traits. Food, predators, weather, and terrain affect survival. Generations evolve over time. Charts show which traits are winning. And I love this one, The Impossible Trip Planner, a planner for absurd fictional vacations such as 48 hours on Mars with 2 children and a $600 budget. This one has destinations, Traveler's budget and constraints, visual itinerary with timeline and a map-like route, budget meter that updates as activities change, random disruptions, meteor shower, missing luggage, alien holiday. And the sixth one is the Audience Boss Battle, a cooperative browser game in which the room fights a giant boss by voting on actions. Four actions per round: attack, defend, head or— heal or gamble. And audience votes can be entered manually or simulated. Animated bus with charging phases, changing phases, damage, status effects, combo system, and final leaderboard. Host can trigger surprise events. So, so really surprised me. How about you guys?

These are so original. I'm so surprised.

Right.

That's what I thought. It's like, wow, I've never seen anything quite like these. But anyway, we're going to choose between 12 different ones. And I'll explain the rules here. You're going to choose by numbers. It's 1, 2 through 12. The number that you vote for and you put it in the chat. Now, here's the real important rule. Only choose one number. No 6 or 3 or 2 or 9, just one number each and that's it. And once you've chosen your number, that's it. Your first choice is what comes up. So it's over to you now to, can you read these all?

Yeah.

Okay, good.

Yes.

Good.

Okay.

If you're driving, can you say the word? Because I think Rosalie is driving. Or say the number.

Okay, maybe she can tell you, Jam, and you can put her in.

10.

She put it—

she said 10. Good. What's 10?

Okay, look, not one of us have picked the same one yet, so let's go, people.

That's what I was looking at. Donnie, can you make a conglomeration of our ideas?

Okay, I'm trying not to look at everybody, so I'm trying not to be influenced.

Right, right.

All right, we have one that's winning so far. Oh, it's really winning now.

Let's see.

No peeking, Donna.

I know, I'm trying not to.

Boring but practical. Selection.

I think we have our winner already.

I think so. I think so.

It's a good one too.

Yeah.

Oh my goodness, that really is, uh, a very simple thing to give Fable 5 or Soul.

Some of the ideas, I was like, oh my gosh, that sounds so jam.

I wanna see the audience boss battle. I'm gonna do that on the side while you guys do this one. I love it.

Are you guys sure about build my elevator pitch? Because we are using heavy duty guys like Fable 5 and Soul. Haiku or ChatGPT basic could build an elevator pitch.

Okay.

Well, let's do something hard.

Yeah, let's do.

One thing, one thing that went through my mind on that, Tony, is since This is much more creative. E.T. was like the vanilla of vanilla historically, right? Now that it's starting to get— and I'm hearing from different quarters as well that it seems like they've gone up two magnitudes in terms of creativity. Now you kind of wonder, something that would be bland otherwise, now could it do something to really come up with a compelling elevator pitch or something that's practical in a day-to-day business use? That was sort of where I was coming from on this, but because otherwise very pedestrian elevator pitch that would put somebody to sleep before the elevator door doors would open.

Yeah. So what do you guys think? Should we keep it simple or see if we can go more powerful? I don't mind either way.

Do you think if we, do you think it'll render fast and we would be able to get 2 in or?

Oh, that's an idea. I think it would run the very, it should run very fast.

Well, actually it would be 3 or 2.0 with Jam off to the side doing number 12, right? That's a freebie. That's a bonus.

Very true.

All right. So let me grab this so that I can put it into— I'm going to have Manus create their prompt. Oh, it doesn't want to let me grab it. Okay. I will tell it instead.

Like, I'm so tired. I couldn't imagine going on a trip right now.

So I'm just gonna go over to Manus wherever he is.

I was thinking about the trip one, but I feel like even real trips I wanna plan are like the impossible trip.

All of 'em are impossible for me.

All of 'em.

They're all impossible.

They're all impossible.

Okay.

Number 2 would be fun on Friday afternoon if nothing pressing was happening. The detective thing might be kind of creative.

Okay, so I'm explaining to Manus here what we're doing, and I have to find the other one so I can— oh, there it is, it's hiding over there. Oh, okay, let me just move it so that I can see it at the same time as Manus. All right, app for tonight, I've chosen as the app tonight Build my elevator pitch. Answer 4 questions, get 3 pitch variants: formal, casual, and funny. Okay, let me get that off the screen. Yeah, over to you to write the perfect Prompt 4, and I'm just going to change a couple of things here. Fable 5. Fable 5. Okay, they've given the same prompt. The audience has no male advantage, but chance. 4 questions, get 3, both Fable-5 and OpenAI Soul. Let's see what he does. Well, he's thinking. How's your boss game going, Cham?

It's going excellent. ChatGPT is already building it out and it said this direction is unusually specific, so I'm skipping design questions and preserving your vanilla front end. I'm just going to start building it now. Okay. And Claude is just saying, I'll build it for you, and it's building it. So they're both creating currently.

Good.

ChatGPT must have heard the comment earlier about being vanilla, and it took it a little hard, and it's repeating it back.

I was wondering that.

Yeah. And when these guys have done this one, we can switch to your screen, Pam, and see what your guys have done. Cool.

Sounds good.

Yeah.

Manus was quick with the prompt. It made a whole— there you go. There's yours.

Okay. Clear enough. Yeah. I just hadn't scrolled down behind everything. It was showing. Okay. We don't need to put that in. Here's the prompt. Okay, what to watch for. Which agent shows a working UI first? Which UI looks more polished and professional? Are the three variants generally distinct and well-written? Copy buttons, loading states, error handling, responsiveness. Which agents ask fewer clarifying questions than just bills? Which agents add something unexpected and delightful? Okay. Oh, wait a minute. This is the prompt. Okay.

That's exactly what I got.

From Manus?

Yeah, formatted just the same.

Oh, okay.

Yeah, that would really be a bummer if you give it different things and you get the same prompt out regardless of what you've chosen. That would not be good.

Okay.

Yeah.

So here we've got Seoul High and Fable High. And let's put in these prompts.

Need to get those trumpets like for horse races, right? As the race begins.

Yeah.

I love it.

I'm also curious to see like which one's faster. Speaking of a horse race.

Yeah. Which one's faster at your end?

Well, I don't know, because they're both just in that mode where they're saying creating. They're both working, right?

Hey, I set you to— I set you to bypass permissions. Damn, it's changed it.

How many of us are bypassing permissions these days?

Okay, I'm gonna stop this so that We don't have an unfair thing. Look, it's set to allow bypass permissions mode.

Huh.

Maybe hit always allow and then it'll go from there. We'll give Claude an extra minute at the end.

Well, we— I stopped him. Oh, I stopped them both. Okay, I'm going to restart him.

I've noticed that whenever you change the permissions and go to a new session, it just carries over those permissions even if you got it set differently.

Claude finished first, FYI.

Okay.

That in and of itself is a good data point so that— so you don't start something to run, walk away, and come back and it's waiting for a setting to be put.

Well, I was wrong. I figured these, these ones would do this in a couple of seconds, but no.

I'm excited to see what it brings back.

Yeah, it heard us all talking and now it's really trying to live up to high expectations.

I love that part. I've con— I've confirmed this is a fresh app and I'm continuing with the setup now.

One clear 4-question flow, instant polished variants, and easy copy reset controls. I'll build this as a hosted interactive artifact, a self-contained single-page app the user can open and use immediately. Let me load the design guidance first, then build it. Okay. And it's used 5, almost 5,000 tokens.

I didn't even think about my tokens.

Shoot, got lost in the moment.

I did, as usual.

And still building too. Oh, now it's at 6,000.

That'll be another good data point. So, Jam, if you're doing number 12, which, who knows, maybe that's theoretically more creative, maybe eats more tokens, who knows, but to compare between these two different prompt types just to see if one is double the tokens of the other.

Oh yeah, that's a good point. Oh my gosh, mine are done and it's creepy because they are exactly the same. And that must have been because Manus's prompt was so specific. I wonder if that'll happen for yours.

Well, last time they were quite different. Oh, okay, this one is finished.

Publishing.

What is ChatGPT? What is Sol doing? Troubleshooting script execution permissions. And this one says published, page not found. Ah, okay. So it created it in a way that was not optimum. Use code browser navigate. Now what are you doing, Soul?

These are supposed to be elevator pitches, not War and Peace novel recreations.

Yeah, but they are making them games, it sounds like. So that might be the differentiator.

Maybe it was too simple.

It doesn't trust us.

So hasn't done anything for quite a while.

Do you want me to show you mine while that builds, or do you want to hold on a second?

Are you still working? Oh, oh no, that's you. That's what you said.

Oh, sorry.

That's okay. Yes, I'm still working. The project is set up and its dependencies are installed. I hit a Windows-specific preview startup issue and I'm switching to the compatible launcher now. After that, I'll finish the interface, verify the pitch generation, and copy controls and publish it. Okay, so it's done. Let me close this and see where we—

interesting, it serves as its own IT help desk support person.

I'm here for that.

I don't know what I just allowed, but I allowed it anyway.

Welcome to my world.

Approve, approve, approve.

Yeah, yeah, I, I could move these to a different monitor if you want to show yours.

Sure.

Okay, give me a second to move them so that I can see what you're doing, and I'll stop sharing. Stop share. Okay, over to you.

All right, of course my link just went away, so hold on and I'll let you know also what's happening with these guys. Okay, so We have— let me get them both side by side like Tony had. We have this one. Let me move you guys. You're not covering it. And then we have this one. Look at— they look exactly the same. It's so crazy.

Oh, yeah.

On the left is—

wow—

ChatGPT, and then this one's on the right.

And they even use the same icon.

Yeah, they used everything the same. Look on the inside and everything. Like it's—

wow.

Wow. Crazy. So much the same, but it kind of should be because we use that Manus prompt, right? So it kind of—

well, last time we used the same prompt for both and got completely different things.

I guess there's a little bit of stuff different here as far as like how it works.

Yeah.

So is saying the app shell is ready. I'm now replacing the starter with the finished experience. And one click. Okay. Claude is still trying to get various things working. Wow. I was so wrong. I thought this would be the simplest thing in the world to do. Does, does your game do what it's supposed to?

Well, I could launch it on Claude and we could all try it because I have that Vercel connector set up. But I can't launch it on ChatGPT for us all to use because I haven't set up my plugins.

What do you have set up, Jam? What is that?

Vercel. It's like I did a little session on it. I'm trying to remember when, but basically Vercel can push your creations in Claude to production and then you can share it and it makes like its own URL and it's a developer URL, so it's not one that looks pretty. It's usually like a random URL.

Right.

But it's at least something shareable that people can see.

And then you can let people play actively, like interactive live right there from it. Is that what's causing the ability for us to—

Yeah.

So let me try it and I'll send it in the chat.

Yeah. I'm glad you were— I'm glad yours worked because I would've suspected that maybe Manus had written the a sabotaging prompt for the other guys because mine, my two are just really just taking, really taking their time.

And yours seems so easy. That's so funny.

I know.

That's what I was gonna say. That's so like life. The thing that we think is right.

Yeah. If we'd asked for an incredibly complex, the one where they had to build a whole city and have emergencies, it probably would've done it in 2 minutes.

Literally.

I think it goes back to Jim's earlier comment where Manus would say, hey bro, let me take care of it. So it's bumbling along the way.

Yeah. Cool.

Well, it says that I have to use Render or Railway because it has to stay on a server that stays on, meaning that like—

Yeah, that's right. Railway is great for that.

Yeah.

It takes quite a bit of setup.

I'm not going to do it because, sorry, I need to save some of my Fable credits. Right. So I'm not doing it for fun, but it was a good, it was a good competition.

Agreed.

Okay. My guys are still working.

So just, Toni, so to Jam's point, when we do create something like this for someone, then once it's done, that's the end of like the, the points and things. It's not going to do that every time we're using it, correct? Like once it renders, right?

Okay. That's one of the things I do like about Manus. It hosts it right there as part of your subscription. You don't have to think about Railway or Vercel or Netlify or any of those things. I wonder how ChatGPT handles it. Do you know, Jam, is, is ChatGPT the same as Claude in—

I was actually just poking around. It looks like it is. I just haven't set it up yet, but it has everything in there that I would need. So man, they really— the more I poke around in Codex here, it looks exactly like quad code.

Yeah. Yeah, and I have them side by side here and they just look the same.

Codex has close to the same or more as code for connectors. Yeah, sounds like Justin, am I right to say that you are liking Codex more for coding projects.

So I have it set up so Claude does the planning, carries out minimal execution. It has to be on Opus or higher to do any of the execution, and then I usually tell it to stop because the amount of things that Codex ends up finding with my execution with Claude is kind of laughable. So I just put it over in Codex and it cleans it up a lot faster.

Interesting. That's awesome.

Do you have anything more to show, or should I bring mine back?

Let me see if there's anything exciting to show. Yeah, I don't think so.

What, what did yours actually do? I didn't actually— I was looking at other things, so I didn't actually see what they do.

Oh, so basically it would be a link I send around and you put in your name and a code and then that brings everybody to the same room and a question will come up and you vote. And if everyone votes the same way, you beat the monster. But if you don't, then the monster goes.

Wow.

Okay.

Could be cool. It might have legs. Who knows?

Yeah.

Watch come Monday, I'll be like launching your personal monster. Knowing me.

You know what, Claude is stuck here. He's trying to solve the fact that dark background is not working properly, and he's spending like 10 minutes on just this thing that is not important. Let me tell him that we don't care.

Good idea.

Unless it's super important, we really don't care whether dark background works or not. Do you want me to share my screen on these guys again?

Sure, let's see what it's made.

I'd love to see it.

Okay, so I got to move you guys off that screen, otherwise you'll be sharing yourselves. Here we go. Okay, okay, one down, two down. Screen 2, share. Okay. Oh, and there again it removes you, so I can't see you. It's not fair. Okay, there's the chat. Okay, so Claude didn't answer. This is amazing me, the fact that something so simple could take so long for heavy— oh, here he says it's built. I'm passing its production build. I'm in the final phase now, checking that no starter content remains. Then publishing the finished updraft app so it's immediately usable from a private link. Okay. How about you, Claude? Or Fable? Fable is using Claude browser for something.

They both have us on the edge of our seats.

Yeah, I'm just beginning to type in the chat that their increased anticipation is a marketing strategy to make its output eventually feel more valuable, worth the wait.

Yeah, these better be good when they arrive.

Well, sometimes it gets hung up on me, so I'll click the little arrow there to see what the agent's doing, and sometimes that gives it a kickstart. I don't know why, but if I look at what it's doing and read where it's at. The time just keeps ticking, but no tokens are used. And if I click on the little arrow there next to what it's doing last, it'll kind of start again, I guess.

You mean here?

Oh, like right there?

Yeah, it's weird.

I'm just gonna ask it, are you still working? I'm not sure if that got in there or not.

No, hit the stop button, then it'll pick that up, and then it'll go again.

Good.

I've had co-workers through the years I had to ask the question, are you still working? Didn't seem evident.

Aha.

Okay, let's see what it say. Okay, here it is. So, your elevator pitch, 3 ways. Answer 4 questions about what you do. Pitch drafts, a formal, a casual, and a funny version of your pitch. Compare them, remix them, and copy the one that sounds like you. So who wants to be the guinea pig? That many, huh?

Come on, people. I will be if you need me to be.

Well, it can be—

likes to do this stuff though.

Pardon?

I feel like Stacy likes to do this stuff usually.

Yeah, where is she?

Oh, she left. Well, she didn't leave. Ah, it—

you can just make it up. It doesn't have to be real.

Stacy looks like she was absorbed by the Aurora Borealis, right?

Yeah. Who are we and what do we do?

Let's be—

oh, I got one in the chat. Hot tub technician.

Perfect.

Okay, what problem do you solve? Fixing hot tubs.

Convert cold tub to hot tub. Yes.

Cold hot tubs to hot tubs. Convert cold tubs to— convert cold tubs, not cold hot tubs, to hot tubs. How do you solve it uniquely?

Prevent electrocution.

That's a result, not a— it's not an action.

You use Flex Seal.

Use Flex Seal.

Okay. Show up with a bottle of champagne for white glove service.

Okay, what's your— what's your call? Okay, you hot tub technicians, what's your call to action?

We'll have your hot tub hot in 20 minutes or less. We'll have your hot tub will be the most hot.

Oh, sorry, say that again.

I'm just being dumb. It's too late for me.

Okay, heat up your nights with hot tubs.

Remember that when I'm speaking, everything you say goes into the prompt. So, we'll have your hot tub up and hot in 20 minutes or less. I think that's close to what you said, right?

Perfect.

Okay, generate my pitches. Okay, formal hot tub technician. The problem we address is clear. Convert cold tubs to hot tubs. Where we differ is in execution. Using Flex Seal, we'll have your hot tub up and hot in 20 minutes or less. Casual. So here's my thing, hot tub technician, you know how convert, you know how convert cold tubs to hot tubs. That doesn't make sense. That's exactly what we fix. Use Flex Seal. Anyway, we'll have your hot tub up and hot in 20 minutes or less. Picture this: convert cold tubs to hot tubs. Tragic, I know. Luckily, Hot Tub Technician and use Flex Seal will have your hot tub up and running in 20 minutes or less. I promise it's less painful than this elevator ride. Both of these ones had some grammatical errors. Wow.

Huh.

Okay, well, that's the one that Claude did. What is Saul doing?

Hello? He was waiting to see what this guy came up with so it wouldn't be any worse.

Yeah. Ah, says the app itself is finished and verified. The private publishing service is still completing the release. I'm watching it through to the live link now.

Okay, so Tony, what's it trying to publish to?

Something that we can— probably something on the web where we could see it, I'm guessing.

It usually does some sort of like HTML, but it's only local to your machine, but we'll see. You never know.

Wow.

I was expecting anything where we would see all these incredible wild games and explosions and animals and plants and all sorts of stuff. And this has been kind of a cold dud.

But it's also real.

Yeah, true. It's a good point. It's real.

I appreciate Stacy's prior optimism. Unfortunately, the systems aren't permitting it.

Oh, I see this one. You can edit the answers. Thank goodness. You can remix. Let's remix the funny one. Yes, that's a real job. Here's the villain of our story. Convert cold tubs to hot tubs. Our superpower? Use Flex Seal. We'll have your hot tub up and hot in 20 minutes or less. Cape optional.

Okay.

Oh, here we go. Hey, I said sign in with Google. There it is. No, what's going on here? Let me make this bigger.

I'm surprised it didn't just populate in.

Me too, me too. I don't have a passkey. Goodness.

Super secret.

Oh no, that part is annoying.

Yeah, get— let me get my phone. Am I trying to log in? Yes, it's me.

I guess I can't get mad at security.

Okay, so Updraft is the name of the— oh, hang on— the name of the tool it built. But that was wild, had to go through all that stuff. Right, signing in and everything. So make your next 30 seconds count. Should we give it the same hot tub technician?

Sure, makes it even.

Yeah, hot tub technician, make cold tubs into hot tubs. I use Flex Seal. What should they do next? Slightly different.

Contact us.

Yeah, actually what he says here is fine. Let's book a 20-minute soak in the tub next week. Same story. Okay, here's formal. I'm hot tub technician. I focus on make— look at this, they both have bad grammar. That should be, I focus on making cold tubs into hot tubs. I wonder why they both have bad grammar. My approach is distinct. I use Flex Seal. Let's book a 20-minute soak in the tub next week. Casual. Hey, I'm hot tub technician. I'm working on a problem that comes all the time. Make cold hot tubs into hot tubs. What I do differently is I use Flex Seal. Let's book a 20-minute soak in the tub next week. Funny, I'm hot tub technician. I tackle make— God, look at that. That's weird. I tackle make cold tubs into hot tubs because apparently wishful thinking wasn't getting it done. My unfair advantage, I use Flex Seal. Let's book a 20-minute soak in the tub next week. Now, I don't see an edit button on this one, which is a pity because we'd have to edit this line because it doesn't make any sense.

Right.

So yeah, so this one looks better, but the Claude one had better controls.

Yeah, it's interesting they took so long to render.

I know.

Hold on, there's a button that said refine my answers.

Where'd you see that?

Upper right.

Oh, okay, thanks. Oh, it's not— it's just refining the questions, unfortunately. Yeah.

Oh, Tony, it took your verbatim. It says make cold tubs in the hot tubs. You were telling it an action and it used your language as the the specific way that it wrote it.

Right. Which is not what it should do.

No, it shouldn't.

No.

And it's funny, I've had apps similar to this that I've created in live session in 30 seconds. So this is pretty weird.

That's really weird.

Yeah.

You know what?

Yeah.

These live sessions this week. Told me that I'm in the right place. I feel like the one about Canva told me that I should keep using ChatGPT for images, and this one's telling me Claude Code is probably the place I should keep building.

Yeah, it is for me. For me, it's Claude Code or Manus. Both of them I like very much. Mm-hmm. But this is a very surprising result from Fable.

Question. I'm just curious. With the new ChatGPT coming out fairly recently, and of course all the people that migrated to Claude know that, right, these LLMs and everything are learning. Do you believe that the new variant of Chat will improve as maybe the pendulum swings and more people go back to it again and it gets more reps with people working with it?

Are you asking Jam or—

Oh, everybody. You, Jam.

I thought it was everybody.

Yeah.

It's interesting question, but what first came to mind is that ChatGPT has been being used the most out of any of the large language models. So it should technically have the most training going into this. So it's really interesting how it was taking what he said verbatim and not using its AI self to rewrite the content in the way it should have been written.

Right.

But both of them, they're almost indistinguishable how error-prone they were in just reading like garbage, right? Certainly not like someone with English as first language. In both of them, I mean.

Let's take a look and see if it was the prompt that was at fault. Build a complete functional web application that helps users craft their perfect elevator pitch by answering 4 simple questions. The application must feature a clean, intuitive input form asking the user 4 specific questions. These questions should prompt the user to define who they are, what they do, what problem they solve, how they solve it uniquely, and what— okay, so far so good. Once the user submits their answers, the application must process the input and generate 3 distinct variants of the elevator pitch. The first variant must be formal, presenting a professional and polished pitch suitable for a boardroom or investor meeting. The second variant must be casual, offering a conversation on a relaxed tone ideal for a networking event or social gathering. The third variant must be funny, in quotes, delivering an engaging and memorable pitch with a touch of humor or wit suitable for breaking the ice. Technical requirements and constraints. Crucially, you must build the application rather than just outputting code snippets. Okay, you're required to use your internal application building capabilities to construct and render a fully working interactive application that can be used immediately. By immediately, it means in 1 hour. You have full autonomy to choose the technical stack that you believe will yield the most robust and responsive single-page application with your— in your environment. The user interface must be modern, clean, and highly usable. This includes employing clear typography, appropriate spacing, and a visually appealing color scheme. The transition between submitting the form and viewing the generated pictures should be smooth utilizing loading states or animations where appropriate. The 3 pitch variant should be displayed clearly, perhaps using cards or tabs, making it easy for the user to compare them or copy their favorite. You are the lead developer and designer. Make intelligent decisions about layout, error handling for empty fields, and the underlying logic to ensure the generated pitch are of high quality. Please begin building the application now. Focusing on delivering a polished end-to-end user experience. I didn't see anything in that prompt that was not fine.

It's missing one thing, the instruction to not make obvious grammatical errors. It's like it needs to be led by the nose in that sense. We didn't make that statement. I say it half tongue in cheek, but that's the only thing I could see missing.

None of us expected that it would do that.

No, no. Well, now we know.

Yeah. Oh, Debbie said, what did you use to create the image for the battle? That was ChatGPT, GPT-2 in WonderReel. And Suzette says, did it make up the edit button? Who, Claude? Claude did, but Sol didn't. Any other questions in the chat that haven't been covered?

I don't see anything. Go ahead, Chris.

There's one, one on the side. For something as— and I think this was divine in, in the selection and the voting process— something as straight ahead and basic as this concept was, and We see not just one, but two of these things mess up and almost identically, and the prompt shouldn't have led to that. You almost wonder if one of the other ones that would have involved more creativity or a little bit more computation or something, right? If there, if that would have gone further off into the weeds, but in more subjective rather than objective ways. I don't know, it just kind of, if it failed, bottom line, if it sort of failed the mission. Something this simple, what would it have done in something that would've been more creative or even more—

I've seen Fable create some incredibly great things and I've seen Soul create some incredibly great things on YouTube, people creating stuff. And I created a Wonder Reel with Fable and it did a great job. So I don't think that would've been the case, Chris.

It's almost like the simplicity confused it.

Could, yeah, could be.

That was going to be my next question. Is this something that almost forcibly diverts you over to the basic model? Say, please don't encumber our advanced models with something that's benign, put it in the lowest level.

Interesting.

Anyway, good experiment. Thank you, Tony. Sometimes you learn what you weren't expecting, and sometimes that can be even greater than the originally intended outcome.

Yeah, for sure, for sure.

Yeah, I don't think you should— I know you're expecting a lot of fun activity and all these apps being built, but honestly, showing what really happens is really important. And it also shows how much editing goes on in a lot of the YouTube videos that we watch, how much people are failing and they're only showing those highlights. So this is really important for us to see. Because when we all go and do it, we're like thinking we're doing it wrong and we get frustrated because this happens. But doing this together shows that this is just normal. So I appreciate you.

Oh, thank you. Yeah, yeah, yeah.

Grateful for this. Our community is so amazing. I'm super grateful.

Yeah, I am too.

What I thought was worth the price of admission is Tony's response to several of these glitches happening and still managing the rest of us here while he's doing it. So hats off to Tony for managing this and still making it learning and enjoyable at the same time.

Oh, well, thank you, Chris.

Yeah, you kept with it. I probably would've killed it off and ran to something else. So thanks for sticking with it and we got it to work. So it was good.

We did. We did. Wow.

Fast.

All right.

Any other questions?

Justin says, I just yell at mine until it gets it right. That doesn't work for me.

I just keep yelling. Never was just, I just keep yelling. It never gets it right, but I just keep yelling.

You gotta be just a little specific though when you yell at it and tell it like where it's being kind of dumb and then it'll kind of fix itself. And if it doesn't, stop it. Until it's still not thinking right.

Yeah, exactly.

Well, I think probably the reason that didn't happen is twofold. One, Tony's a gentleman, and two, I think we were all bewildered at what we saw with all these— it's like somebody taking a Ferrari to drive down the street to the 7-Eleven, right? It's something's— something went wrong here and it wasn't our fault.

Yeah. All right. Well, also, Tony, I can't wait for this recording because I thought there were some really good ideas in the beginning. So I want to look at those again and we'll talk Friday, Caffeine and Convos, everyone. So go play with Codex and ChatGPT and come back and tell us what happened with your internal battles that you're having with these large language models.

Do you want a screenshot?

I'll wait for the recording. Maybe you could send the deck along with it and I'll upload it in the classroom.

Okay.

All right.

That could be our new thing, Jim, is if anybody goes off and plays with it and they get equally broken, results, do screenshots and post it in the community for good chuckles.

That can be our bloopers.

Really good idea.

Exactly.

All right guys, have a great evening.

Thank you. Awesome job again, Tony. Thank you.

Thank you.

Thank you.

Bye everybody. Bye.