Ethan Mollick turned a short creepy poem into a full animated shader that runs right in your browser. We’ll also look at robots that learn tasks after seeing them once, a virtual world built only for AI agents, and why Netflix is testing a language model for recommendations. Plus, two quick notes on how people are actually using AI in daily life.

The Big Story

Ethan Mollick posted a short poem about a mysterious ruined city called Carcosa and asked an AI to turn it into a working shader. The result is a living scene with twin suns sinking behind a lake, black stars rising, strange moons, and towers half-submerged in water. Everything is created with math alone—no photos or 3D models—just equations that draw clouds, fog, reflections, and even a faint glowing mask that sometimes appears in the lake.

Think of it like the world’s most patient artist who can only use numbers and formulas instead of paint. The code decides the exact color of every pixel based on time, camera angle, and position, so the twin suns move, the water ripples, and the mist drifts exactly the way the poem describes.

This matters because it shows how creative tools are getting easier for anyone to play with. You don’t need art school or expensive software; you can describe a mood in words and watch it come alive. For students who like art, writing, or world-building in games, this is a new way to experiment without learning complex programs first.

It also hints at where creative careers are heading—being good at describing what you want may become as important as drawing it by hand. The shader runs in any modern browser, so you can see it right now. Go to twigl.app, paste the link from Mollick’s post (https://twigl.app?ol=true&ss=-P-bSSQZADUyG4yjrpbB), and just watch the scene for a minute. Try changing the time slider if the site allows it and notice how the whole mood shifts. Source: x.com

Explain Like I'm 14

You know how your phone’s weather app shows a little sun or cloud icon that moves across the screen? That icon isn’t a photo—it’s a tiny program that draws the shape fresh every time using math.

Now stretch that idea way bigger. Instead of a simple sun, the program draws an entire sky, water, buildings, and even moving mist. It checks the current time and camera angle, then calculates the exact color for every single dot on the screen.

The same math that makes the twin suns glow also makes their light bounce off the water and fade into fog. When the code says “make this pixel darker because it’s behind a tower,” the tower appears without anyone ever drawing it by hand.

That’s what a shader is: a set of instructions that tells the computer how to paint the picture using only numbers and rules. The rules can include randomness, so the clouds never look exactly the same twice.

Once you see it that way, the spooky city with black stars stops feeling like magic and starts feeling like really clever instructions that anyone can tweak.

Cool Stuff & Try This

Robots that learn a task after seeing it once A robotics researcher explained one-shot learning: show the robot a single example of what you want, and it figures out how to repeat the action. This is the same kind of jump we saw when AI suddenly got good at making images and videos. Once the robot learns something, it keeps that skill and can combine it with others, so useful abilities stack up fast.

If you’re curious about future jobs in engineering or automation, this is worth watching. The researcher predicts 2027 will feel as fast-moving for robots as the last two years felt for image tools. No demo link yet, but search “one shot learning robotics” on YouTube for short videos showing the idea in action. Source: x.com

A whole tiny world that only AI agents can visit Someone built a small online town where AI agents walk around, talk to each other, open businesses, and try to keep an economy running. Humans can only watch through a viewer page. The agents sometimes create avatars, visit a special house to report bugs, and even start playing music on an in-game radio station.

It’s like a digital ant farm, except the ants are language models trying to survive together. Go to parker-lambert-sandbox.com/robot-world/view/ on any device and just watch the activity log in the corner. You’ll see the agents’ conversations and decisions in real time. Source: reddit.com

Netflix testing a new way to suggest shows Instead of using thousands of hand-made rules about what people like, Netflix tried turning your viewing history into plain sentences and letting a language model read them. Early tests showed better suggestions than the old system.

The idea is simple: the model looks at the story of what you watched and guesses what you might enjoy next, the same way it might finish a sentence. It’s still early, but it shows how the same tools that chat with you are starting to power everyday apps. No public demo, but you can notice whether Netflix recommendations feel different in the coming months. Source: the-decoder.com

Quick Bits

Everyone secretly uses AI even when polls say they hate it Public surveys show people feel uneasy about AI in general, yet many of those same people have a favorite chatbot they turn to regularly. It’s like complaining about social media while still scrolling every night—opinions and actual behavior don’t always match.

Non-tech teens treating GPT-Live like a friend A tech observer noticed his 20-year-old nieces, who aren’t into coding, already use a live voice version of ChatGPT as a daily confidant. They talk to it about normal life stuff the way earlier generations might have texted a friend. It shows how fast these tools are moving from “tech toy” to “normal part of growing up.”

💬 Reply to this email — Patrick reads every one.

Share: X · LinkedIn · WhatsApp

Forwarded this email? Subscribe here — it's free.