Welcome to the AI For Humans newsletter!

OpenAI has been teasing a new flagship model for a few weeks now.

You might be asking yourself: Wait, didn’t OpenAI just release a new model?

Yes. They did.

GPT-5.4 came out literally a month ago.

Alas, this is the way of AI now. You might’ve heard… it’s moving faster.

While we’re not sure when we’ll get the new SPUD model, what we did get over the weekend were some leaks of their new image model (currently referred to as GPT-Image-2) and whooo doggy, it’s good.

Nothing confirmed so far, but a few new image models appeared on the Arena.AI website (where you, the user, are asked to judge which of two anonymous models are better). And a few random people were saying they showed up in ChatGPT.

We’ll get into a more examples below, but one of the more striking things is just how solved AI images look right now… and the implications of that for AI video and, ultimately, AI worlds are somewhat scary.

Please support AI For Humans by learning about our sponsors below:

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

As AI images have improved over the last few years, something that’s started to differentiate newer, better models is weirdly… more defects.

Real-life photos or images rarely look perfect and AI images became easy to spot because they often looked too good.

The issue was getting to what looking real actually meant… like those imperceptible imperfections on human faces and environments.

In the example above, the thing that’s most striking to me is the shadow of the person taking the picture on the lower half of the woman’s body.

This is the kind of strange detail that AI imaging wouldn’t really be able to illustrate before (there’s a lot of stuff to think about) and, without even noticing, adds an immense amount of realism to the image.

And, maybe even crazier, it seems like this new model has the ability to nearly perfectly recreate modern web screen shots and a LOT of text.

All three of the examples below are remarkable but click through and take a look at that YouTube “screenshot” and count the number of gibberish words.

Ok, I’ll save you the time. It’s zero.

Yes, the Mr. Beast and Andrew Huberman thumbnails look slightly off but that is a full fake webpage that I’d argue looks very, very close to the real thing.

That is both astounding and… kind of frightening.

Love this newsletter? Forward it to one curious friend. They can join in one click.

It’s not like we haven’t been here before (we seem to have mostly weathered the DeepFake issue) but as the output of AI imaging gets closer and closer to the real world, it’s worth thinking through a few of the larger implications.

First and foremost, AI images are always the first stage.

I just chatted with Puck’s Julia Alexander on her podcast about how Midjourney v1 was one of the first times I felt like AI media was going to be a disruptive force.

Even then, my first thoughts went towards video and fairly quickly after that, we got the infamous Will Smith spaghetti eating clip which many, many people laughed at.

Then, just a month ago, Seedance 2.0 gave us this:

And, if you extrapolate out even slightly further, you start to see something even more crazy because what’s coming next after video is fully interactive AI environments aka AI world models.

These are just at that nascent stage of generation (Google’s Genie models are a good example) but they’re coming and they’ll follow the same path.

Take a close look at these new leaks and you can start to imagine a world where not only will you get video clips as real and detailed about this but about a world that’s entirely generated that you can step into and… live in?

We’ve said it many times before… we’re in for a few very weird years.

See y’all next week for a new episode!

-Gavin

Last Friday’s AI For Humans: The 1.8B Vibecoded Start Up👇

It’s hard to believe it’s been more than full year since the last big ‘vibe game dev’ moment but here we are.

Thankfully, Pieter Levels (@levelsio on X), creator of the mini-flight sim that kicked off the initial vibe game rage, is back with a new Vibe Game Jam for 2026.

Sure you can win up to $20,000 but if you’ve been thinking about a weird fun game idea, this is your chance to give yourself a finite window to ship it.

Kind of like NaNoWriMo, it’s not about the end product as much as it is about pushing to get something done and out the door.