Roundup: Bootstrapping Growth Without Cash — Free Readers Issue

Hey Prompt Entrepreneur,

Welp. Quiet week. Nothing to report.

Oh wait…there was a little bit of news from OpenAI I suppose!

So! It’s here. GPT-5 is finally in our grubby little hands.

And…it’s OK? S’alright.

I did a livestream alongside the OpenAI livestream. A simulcast? Simulstream? Not sure what the cools kids say.

Here’s a 3 min vid with some thoughts:

iamkylebalmer

6,501 followers

View more on Instagram

Let’s quickly run through it though.

First up, the presentation itself was pretty amateur. But…I kinda love that. These are engineers. They aren’t swish presenters. They stumble and fumble over their lines. Mess up product demos. Never know where to put their hands. It’s endearing.

It also means what we see is much more “real” compared to a flash Google or Microsoft demo where we get wowed with pre-shot professional video and then left somewhat cold when we finally get the tools in our hands.

Second, the main thrust of the presentation was “it’s better”. Better at writing. Better at coding. Better at any problem you can throw at it.

It’s a weak hook unfortunately! A bit like recent Apple presentations where they show us an iPhone with (wow!) a better camera. Who cares?

We are approaching a point where ALL of the models are really damn good. So showing something that blows peoples’ socks of is genuinely hard.

Sure, they can show us benchmarks. Sure, they can tell us it’s smarter. But it’s all kinda meaningless until we get it in our hands and play around with it.

Third, I got access the day of the release. First in Cursor then on phone. And then a day later (Friday) on web. It’s a staggered rollout but none of the delays like we saw with Sora and advanced voice mode. You’ll get it shortly if you don’t.

I immediately fired it up in Cursor and threw some tricky coding tasks at it. And it absolutely smashed them out of the park. It may become my main coding model. I still need to test more but it’s very good.

I think OpenAI knew this was their Ace card as they spent almost 50% of their presentation talking about coding. I’m not surprised.

In terms of “general” usage it’s less impressive. To me (and this is subjective) it feels no better than what we had a few days ago. It’s still very good. It just hasn’t leaped light years ahead. And nor should we expect it to!

Other people are having, hmm, different experiences. They HATE it.

This is wrapped up in the fourth point: the unified model.

Now when you go into ChatGPT you’ll see 4o, o3, 4.1, 4.5 etc. are all gone. They’ve been deprecated. They are dead. And buried. Without any ceremony. OpenAI took them round back, shot them in the head and dumped them into shallow graves…and replaced them with GPT-5. The King is dead, long live the King.

This has pissed off an awful lot of people. Want to see some anger and fury whilst sipping your Saturday morning coffee? Go check out the ChatGPT subreddit. You’d have thought Sam Altman had kidnapped their children the way they are going on!

The root of this anger is that the model is now “stupid”. Well…it’s not. The model is solid.

The problem is that because there is now a unified model it chooses how to deal with a prompt. Sometimes it’ll route your prompt to it’s more advanced reasoning. Sometimes it’ll route to (presumably) something like 5-mini or 5-nano in the background. These are the lobotomised cousins. And they are cheaper to run, hence the routing.

The problem is sometimes a query goes to the state of the art model. Sometimes to Flowers of Algernon. This leads to unpredictability. Which is driving people mad.

Previously we could at least choose to use an advanced reasoning model like o3 when we knew we had a more complex problem to tackle. Now it’s a crap shoot. Sure we can say things like “think hard” about this. But there remains a level of opacity.

I wouldn’t be surprised if OpenAI step back from their unified model and add in an “advanced mode” or similar that gives more control. Basically adding back in a dropdown menu like we used to have.

Fifth, and an important reminder. For the vast majority of ChatGPT’s 700 million users this is all inconsequential. They’ll wake up and use ChatGPT like usual. They may notice that the main screen is colourful now. Cool. Maybe they’ll see welcome messages. OK, sure.

But most people do not care about these intricacies. And very likely they’ll be perfectly happy with the new model. Pleasantly surprised in fact when it triggers (unbeknownst to them) into reasoning mode and gives them a much more detailed answer.

So: the devs will be happy (I think). And they are using the API so model choice is up to them. It’s another tool in their stack. And the general (700m!) users will be happy.

It’s the weird middle where a lot of us hang out that is furious! But remember we’re in a bubble!

These are my (rambling!) first impressions. I’ll form up more cogent opinions once I’ve used it more.

But for now?

S’alright.

Keep prompting,

Kyle

Follow for more X, TikTok or IG.