Tomorrow Tuesday — August 2026 (Issue #4)
Welcome to another issue of Tomorrow Tuesday, where we cover the revolution running through film, games, music and the creative industries.
A newsletter by SOUTS, where we explore what happens when artificial intelligence stops being just a tool and starts changing how we create and consume stories.
Do We Actually Prefer Stories Written by AI?
This week’s newsletter starts with a new study from Villanova University that brings up something uncomfortable: readers prefer short stories written with artificial intelligence over ones written by humans.
More than 2,000 adults between 18 and 81 took part in the study. The researchers used three short stories written by human authors published in well-known outlets, and three written by ChatGPT.
Participants were given a short story and asked whether it was written by a human or by an artificial intelligence. Then they rated it on quality and on how engaging they found it.
The results
The stories written with artificial intelligence got better ratings for quality and engagement than the ones written by humans. And they scored even better when participants were told they were written by humans.
Participants’ ability to identify AI-written text was just under 50%, which is basically a coin flip. In other words: we are very bad at spotting text written with artificial intelligence.
Participants who said they used artificial intelligence more were considerably better at identifying the texts.
Participants who said they had more experience with literature were not any better at telling artificial intelligence apart from human writing.
My takeaways
I think there’s an obvious bias here: we humans like reading stories made by humans. Just knowing (or believing we know) that a story was made by a human makes us like the story more.
It makes sense.
It matters to be familiar with these technologies because, as the study shows, the more we know their quirks, the better we get at spotting when they’re being used, and when they’re being used badly.
And if you don’t feel like actually using artificial intelligence, you can just read LinkedIn, which is pretty much the same thing.
This experiment reveals something even deeper than the AI part. The texts chosen as human literature weren’t picked at random: they were pulled from short stories published in reputable outlets by reputable authors. Human literature, and especially ‘award-winning’ human literature, tends to ask more of the reader: it’s subtler, more complex, harder to process.
Literature written by artificial intelligence, on the other hand, tends to be very clear, direct, not very subtle, easy to process. For example, LLMs usually spell out the theme of a short story right at the start, instead of leaving you, the reader, to infer it on your own through the characters.
We live in a moment where our attention span keeps dropping, where fast, immediate stimulation is taking over our lives. My hunch is that this study is, above all, a symptom of that.
Why the study came out this way, we don’t know. That’s the fun of studies. We can speculate all we want, but the reality is that participants couldn’t identify AI-written texts and those texts were perceived as higher quality than the ones written by professional human authors.
Hell Grind: What’s the Deal?
Higgsfield released the film Hell Grind on YouTube, a Higgsfield original. As I write this it has 220,000 views. The pitch is that it was made by a team of 15 people in 14 days for less than USD 500,000.
The film had been shown at ‘Cannes’ (in quotes because it wasn’t presented at the official festival), but now it’s making noise again after being published on YouTube.
Main disclaimer: I didn’t watch the whole thing. I watched the first 10 minutes and some scattered fragments.
With that, I have two opinions:
- More than a film, it’s a 95-minute ad for what AI (and Higgsfield) can do.
- Even so, I want to watch the whole thing.
It has technical quality, it has a story, it has reasonably interesting characters. It’s the state of the art in AI filmmaking and it’s incredible that we’re at this point. I’ll give you my review later.
Why does it personally keep me at such a distance? The film has been available for 6 days and it’s only now, to write this newsletter, that I sit down to watch it. It’s not a lack of time: over the weekend I watched two movies (Sleepless in Seattle and The Phoenician Scheme). Let me think out loud about my reasons.
- The marketing. I HATE these Higgsfield messages of ‘We made an 800,000,000-dollar movie in a weekend’. I think it does damage to absolutely everyone.
- These films that end up being technology demos make me feel like I’m watching something with no substance at all. Even the genre is picked to camouflage that lack of substance: always fantasy, sci-fi, monsters, aliens.
- And finally, it’s Higgsfield. I’ve already talked about Higgsfield and the shitty practices they have. Tim from Theoretically Media has a detailed video on one of those practices, and there are others here and here.
Look, this film isn’t going to win any awards and we’re going to forget about it in a couple of weeks. But setting my emotions aside, it’s a big step in the direction we want cinema and AI to go: more original stories.
The best thing about it, and here I applaud Higgsfield: they published the entire process of the film on their site.
From character creation to the brief, the creation of every asset, the scenes. You can go in and really understand how a film gets organized and made. That is very valuable!
What Does It Mean for LATAM?
Two years ago, when I started the studio that is now Souts, I made a prediction: AI in video is going to improve month after month, the barrier to entry is going to drop, AI production companies are going to pop up from under every rock, AI video is going to become a commodity, and we’re going to end up in a race to the bottom.
Congratulations, just turn on the TV during any ad break and you’ll see all of that came true. It wasn’t that risky a prediction. The future is already here, and it’s boring.
This semi-magical technology showed up out of the sky promising you can create quality audiovisual content for a fraction of the price.
I get that everyone’s first impulse is to go for what looks like easy money.
But there’s another path. If you no longer need enormous infrastructure to produce your content. And you no longer need enormous infrastructure to distribute your content.
Why aren’t more companies, more agencies, more production houses popping up in Latin America dedicated to creating original fiction, original intellectual property, original characters, powered by all these technologies and this new economy?
Let’s break it all.
A New Image Model Is Coming
We close this newsletter with a preview of an image model that’s coming out in the next week or two and that will be the image model with the biggest impact on everyday life over the next month. It comes with a bonus explanation of why Google’s image model has a name as dumb as Nano Banana.
Here’s the thing: every big AI lab wants their next model to be the best when it launches. There are a lot of steps for that, all inside the lab; but that doesn’t tell them much about how normal people will perceive that model. So there are these siteswhose whole job is pitting model against model. Users drop in a prompt, that prompt runs on two models at once, and they pick the better result. The user doesn’t know which models are competing because they all have a code name so nothing leaks.
This is one way labs actually find out whether normal people like or don’t like the images their model generates.
This week a new model showed up in the top spots of that site’s leaderboard: some “mona-lisa-1”. The first suspicion came from how fast this model started climbing the ranking, which suggested it came from a major lab.
Digging out provenance data in the metadata of the images it produced, people started finding traces that the origin is indeed OpenAI.
The Good
Better 2D consistency. Characters are more consistent, anime and illustration styles work better, fewer anatomy errors.
Less plastic look. GPT-Image-2 famously produced people with a not-quite-realistic look: skin too smooth, light a bit artificial, too much detail. Mona Lisa 1 seems to improve this, though not fully solve it.
Better at non-photorealistic styles. The big leap seems to be in better text generation and illustration, concept art and stylized results.
The Not So Good
It doesn’t look like a brutal jump. The consensus is that it’s an incremental improvement rather than something completely revolutionary.
There isn’t a huge amount of progress on realism. GPT Image 2 was already a pretty good model at everything realism-related. This one seems to be on par. It does improve the render and the plastic looks, but the advance doesn’t go much further than that.
So why do I say this model will have the biggest impact on everyday life? Because ChatGPT is the most used language model in the world and its image system is one of the most used because of how practical it is. We already saw, when GPT-Image 1 came out, that all the ‘low quality’ content started to have a very similar look. You’d walk into a corner store and see the signs with that characteristic yellowish GPT style. When GPT-Image-2 came out, we started seeing images everywhere with heavily outlined typography and extreme detail.
And now, when this Mona Lisa 1 comes out, we’ll see what they end up calling it. Once it actually ships, we’re going to start seeing on every sign around us the artifacts we’ll gradually discover in this Mona Lisa 1. Because we already know it’s going to have its quirks, every model in the world does, and they’re going to get more and more obvious.
This was Tomorrow Tuesday, the weekly newsletter. If you like it, subscribe!
At SOUTSwe try to always stay on top of what’s happening, and every time there’s something relevant we’ll bring it to you in this newsletter.
See you next Tomorrow Tuesday.
Yves
