Inside GPT‑4.5: The Surprising AI Upgrade Marketers Can’t Ignore.

We have been under an avalanche of AI news. We've got new models from Anthropic, from Grok, from everyone out there. But I want to tell you and Kieran, I want to tell you about our favorite AI model of the week. And that is GPT 4.5, just dropped from OpenAI. It's got the vibes, it looks amazing, it works amazing. And we are less than 24 hours with this model and we love it. The big picture here though is that GPT 4.5 has the best vibes of any model. Kieran, you're all in the this is your vibe AI era. Why does GPT 4.5 have the vibes?

We have all the intelligence we need. We want to do certain tasks with the AI. So how do we choose the one we want to hang out with? It's the one you vibe with, right? No different from how you decide what colleague you want to hang out with.

But what's interesting, Kieran, is one of the tests that I keep seeing is like the hand drawn test. Do you know what I'm talking about, where you ask a model to replicate a drawing?

Yeah. The creative part, I think is a story here.

So, yeah, so if you look at the new ChatGPT 4.5 GPT 4.5 model, it's just very creative and very high taste. And so Aidan here on Twitter asked it to redraw this hand drawing and it asked 4.5, the new 4.5 model, the newer Sonnet and the Grok 3 model. I don't know if you've seen this, Karen. This is the newer Sonnet. So. So Claude Sonnet 3.7. This is how it redrew that model. Not very good. Right. This is how Grok3 did it. Much better. But like, it kind of looks like reindeers, like hanging out in the circle. And this is the chat GPT 4.5. So if you remember. Oh, triangles with pointed together with stick figures on them. It's not all the way there, but man, it was much, much closer than everything else.

It's a. It's much better.

One of the things I want to show people right off the top for the show today is it is really good at writing.

Yeah, it's really good at writing.

Might be the best model at writing now. And so it's like ChatGPT, this is their clod killer. Let's be real, right? This is what it is. You and I chatted offline and you're like, they realize that vibes matter in these models. And that Vibe is Basically the user experience of these LLMs. And so Ben said he's been testing out 4.5 for the past few weeks and it's really good at writing. I thought this was entering. This is literally the midjourney moment for writing. So where midjourney came along and made image creation really good. This is the same for writing. And it says his prompt was write me a four sentence terrifying story with an insanely surprising ending. Something that no one has ever heard before. No one would ever predict something Stephen King might write, but make it simple, approachable, make it a little vulgar too. Here's the story. Late at night, alone in my shitty apartment, I heard whispers sliding out from under the fridge. Soft wet voices promising they'd crawl inside me, hollow me out and wear me like pajamas. Terrified, I yanked open my freezer door expecting rats or roaches, but saw only ancient freezer burned pizza rolls crusted with ice and regret. Suddenly the rolls moved, splitting open to reveal tiny mouths lined with needle like teeth whispering in hungry unison, We've waited for you, asshole. Before I could scream, they erupted from the freezer in a greasy swarm, burrowing into my flesh, forcing me to realize, oh fuck. My midnight cravings had finally gotten their revenge. That's really good.

Yeah, it's really good. I think this is a that's really creative look.

We all love ChatGPT. You've probably used it before and if you haven't, stick around because you're going to want to listen to this. You know, I've used AI a ton. I use ChatGPT for research, for writing. I use Claude to put guides together for whether I'm taking a trip or I'm working on a project at work. It's awesome. Well, I've got something that will make ChatGPT an even better experience for you. We've got this killer chatgpt bundle that will take your prompting to the next level. It has over 100 prompts and a step by step guide to integrate it across your workflow. It is a total game changer if you want it. I'll drop a link in the description below. Now let's get back to today's show.

The big takeaway here for people is this is a really creative model. It really has the. It's the era of like vibe, AIs, how, how do, how, what are the vibes? And so I would equate it to when people ask me why did I use Claude? And I kind of tried to articulate it just felt more like someone I would want to hang out with that is the Vibes. And so from using GPT 4.5 this morning and from even listening to their launch video, they talked a lot about the Vibes and the creativity and the. And I think that is really important. So I used it for write in this morning and it really is the first model that I think is equivalent if not better. And I can't say it's better because I haven't used it enough to Sonnet 3.5, but it's definitely on par. From what I have tried so far.

I think it's on par or better. I think what it is and what makes it so remarkable for folks who don't know the full background, this is a very smart model, but it's not a reasoning model and it's different than those reasoning models that take lots of times to think. It's actually really good at thinking quickly and getting things right in one shot. I don't know if you saw this, Kieran, but this was wild. Did you see this?

I did not see this.

It's the only model that this doctor has used so far that properly identified an ultrasound. Right. That this was a. This was a entopic pregnancy and all the other models just thought it was a normal pregnancy and he was talking about how remarkable that breakdown is in that like it didn't do any reasoning, it didn't have to think. It actually got it right on the very first try. Which is wild.

Yeah.

But in line with this tweet from Aaron Levy Kieran, where they've been text testing 4.5 versus 4 oh at box with a bunch of data sets and they found a 19% improvement in single. Single shot or first attempt tasks. Like it's way better at getting the right answer on the first time.

First shot it has better. It understands the world better. And the point that you made, just to reemphasize the point you made, it's a new base model. So O3 were built on top of its. Its reasoning engine, whereas this is a new base model which means that there are likely going to be new reasoning models built on top of this as the base. And so why are they doing that? I think they are moving from a world where AI is your like bland assistant and you kind of go to it when you need a task to your conversational best friend and they will build future reasoning and thinking models on top of this and it will start to feel like your kind of kind of friend. Now I do wonder. One of the things I do wonder about is if you are Listening to this like the immediate takeaway if you can, if you're lucky enough to be able to afford it, is wow, I'm going to go try it for creative tasks. One of the comparables here is something I have heard about Cloudsonnet 3.5 is it's better at logic and coding because it's their first thinking and reasoning model and far worse for creative tasks. And so I do wonder the more you then start to layer on better code and better logic on top of these models, can you have one or the other? And that's probably why like GPT5, one of the other takeaways here is OpenAI have just made this so confusing, right? It's a big drop down box and you have no idea what real model to use. And for the average user they just want to be able to say something and the AI to like figure out does it need reasoning? Does it not read reasoning? Does it need creativity? Does it not need creativity? That's why I think this is a step towards GPT5 which will be a text box and tech. The text box will take your input and it will decide in the background is this a creative task? Is this a reasoning task? Is this a deep research task? Is this just give you back the answer as quick as possible task? And I think this is a further step towards that.

I couldn't agree more. And I think what's interesting here is that a reminder to everybody, this is a very expensive model to run. I think the API calls are like $75. I forget like it's insanely expensive. It's only available to the $200 a month pro subscriber class. Sam Altman, CEO of OpenAI said they ran out of GPUs. They're going to add GPUs. It's very GPU intensive and this is going to roll out to plus users I think in the next week or so. But all those GPUs Kieran it is like quality of taste is really good. This tweet really got me which was basically give me one truly deep novel out of distribution and mind blowing simple insight about humans that only very few or none of us are aware of. Humans never genuinely pursue happiness. They only pursue relief from uncertainty. Happiness emerges momentarily as a byproduct whenever uncertainty briefly disappears. And there's a whole whole lot of things there. But like that's why has more creative.

I put I want to show you a quick actually I think this was one of my best summaries for how I felt about the Quick. I don't know if it resonates with you, just the time I spent with it. So this is a pretty good tweet from Elvis Shout Out. I've taken some of his courses at Dart AI and I kind of think he matched pretty well what I thought about 4.5 in that it's a top model for writing. So if you take nothing away from this, it's wow, you actually have a model again if you're able to afford at its current cost. But there's a great model now that is a really great creative write in assistant and write in thought partner. And I've always used Claude. I can see me really starting to compete that usage with GPT5. It's great at creative tasks. So creative thinking. This probably is the model for marketers, right? Like really creating.

That's what the show's about is we're showing you.

This is the model for marketers. It's like a. It's a great creative thought partner. It is a fun personality. This is the Vibes. I think this is the introduction of the Vibe LLM era, where these models realize if you are going to have a slew of different models available, how will humans choose? Because for the. You talk. You and I talked about this earlier in a different meeting. The average user doesn't need more and more models. Why don't you actually make that point? Because you made a great point that I think is really important here, which is the average user needs Vibes and all of the things around the model to make it useful.

First of all, I love that I picked a profession that it's AI anachronistic Thing is Vibe AI, like marketers are all about the Vibe AI and it makes me really, really happy. I kind of want to make like Vibe AI swag. Like, don't you want like a Vibe AI, like hoodie or something?

Oh, darn. I'm going to call my newsletter. I was. I'm relaunching my newsletter. I've just thought of the name Vibe AI.

Vibe AI. We should. You should go look right now. Vibe AI is available.

I'm looking actually.

And the point I was making is like these models, they're really good. Like 4.5 is amazing. Like if they didn't make it, any improvements on it for the next six to 12 months, I'd probably be very happy. The thing I actually need now is I need it more integrated with my other work tools. I need longer contacts, windows. I need it to be faster and cheaper.

I don't need bigger Context Windows.

Yeah, I don't need these crazy, crazy high intelligence models. I can do most of the things I need to do with what we already have. And I know we're on a race to 10x from here and that is insane to me.

The average user does not need more intelligence. They need it to be more functional and integrated into what they want to do. And they need to cool vibes. I do agree they need to cool vibes because ultimately what they're trying to do is have a singular model that you want to hang out with and do work with. And for that, the vibes do matter.

GPT 4.5 and Claude Sonnet 3.7 are now the two most important models for marketers on the planet. That's what you need to know. And if you are using AI for writing, for creative output, for creative brainstorming, for deep research leading into the. Into the. Into those outputs, 4.5 is probably going to become your favorite model for that over time. I've been with it for less than 24 hours now and I have been doing some things in it and I find that to be very true for me and that it is probably the peak model for marketers. You know, this week we'll see what happens next week or whatever, whatever's going to come out. But it's that. That's the takeaway. It's expensive. It is the Hermes handbag of AI models. It looks great, it works great. Super hard to get, it's super expensive. But you're gonna be really happy if you.

But it's worth the vibes. It's got all the vibes.

It's got all the vibes.

It's got all the cool vibes.

All right, all right. That was GPT 4.5. We'll be back real soon with another episode. Marketing gets the grain.

This data is wrong every freaking time.

Have you heard of HubSpot? HubSpot is a CRM platform where everything is fully integrated.

Whoa.

I can see the client's whole history. Calls, support tickets, emails. And here's a task from three days ago. I totally missed HubSpot grow better.