This time, I had Fable built me a casual Van Gogh city building game that I faked in an AI video last year. The key mechanic the AI came up with is painting the landscape with big brushstrokes & an environment that evolves with weather and seasons. Chill and pretty. Play it: the-sower.netlify.app

Yes, the AIs were given a cybersecurity challenge, with internet access enabled and safety filters disabled But the extent to which Mythos 5 pursued its mission (fake identities, social engineering, inserting malicious code into a real open-source project) is notable www.aisi.gov.uk/blog/inciden...

Given today, it is surprising how daring Microsoft & Google were initially with AI. Microsoft released GPT-4 before OpenAI, didn't back down after Sydney & got Copilot to market quickly (the 1st professional AI tool) Google dealt with blowback about Bard and AI Overview and invented deep research.

An unexpectedly stress-reducing use of Codex/Code is just fixing problems with my various Windows machines: weird driver issues, game incompatibilities, even just tiny stuff that used to annoy me (why does the program that I set to run at startup not run at startup?). Hours saved. Annoying hours.

Sure, Threejs in webpages are neat but, Codex: "you have access to Blender & Unity. I want you to make a new game, with full assets, in which you play as an otter who can get into mech suits shaped like animals and that are critical to game play" Took over my computer, built assets & gave me this

And yet they still stink at good long-form fiction.

I continue to think that a lack of verifiable answers in many fields is a real issue for LLMs but not as big a problem as it sometimes is made out to be. As models are improving at formal domains, they also are Improving at lots of other less-verifiable domains as well, though jaggedness remains.

OpenAI announces 10 discoveries from their next model. Observations:: 1) AI is getting very good at math 2) Two years ago LLMs failed at basic math 3) This cost less than $2000 in current API fees 4) OpenAI is focusing on announcing benefits, not just risks, of new models openai.com/index/ten-ad...

One big result in our study at Procter & Gamble was that AI blurred the lines between jobs. Now OpenAI has a similar finding Organizational boundaries are becoming porous, the walls thinning. Companies are going to need to think about division of labor in a new way, things are getting chaotic now.

Continuing a trend, I had Fable build a working Rothko-inspired city builder based on the fake AI video I created a year ago. This time, the unique mechanic the AI developed is that you play with margins between colors & forms to grow your city Its worth a few moments: threshold-city.netlify.app

I thought sycophantic models were bad but then they all got nit-picky instead.

Ted Underwood@tedunderwood.com · 6d ago

man, I miss sycophantic models these ones, if you thank them for answering a question, they're like "not so fast! — I answered the question you posed, but didn't address possible blind spots in the question framing itself"

Managing AI requires new interfaces. These are the physical things I have been trying out to control Code. Given how useful voice mode is, the Teenage Engineering Ting walky-talky has been a surprising favorite. The Codex Micro is beautiful & fun, but displays too little info vs the Stream Deck

I gave Flux 3 the penultimate lines of Eliot's The Wasteland, which involve both switches in tone and in language. I had it read by a modern Fisher King as a city decays behind him. The results were surprisingly good.

The Executive Director of my Lab, Laura Zarrow, a former art school dean, has been teaching students to work critically and effectively with AI using a studio structure derived from art education. She wrote up a guide to the methods she developed: gail.wharton.upenn.edu/gen-ai-studi...

Our lab just released our AI Behavioral Observatory open source. It lets you run statistically valid tests on how AI behavior changes under various types of prompts. We have been using it for our own studies & I think it could help others do similar work. gail.wharton.upenn.edu/research-and...

From building an AI research tool to prompting the research itself

gail.wharton.upenn.edu

Fable is amazing but needs to stop talking like someone who has read only pulp fantasy: "I have shown you the way but you must open the door. The map exists but the path is yours. The atlas of your instinct — every fact must first know itself" Please, just make the infographic about cheese I wanted

Flux 3 is pretty darn impressive. This is what it produced with the prompt: "tracking shot that follows a female astronaut with her helmet open as she walks through a regency dance in a traditional manor, with a mural on the wall painted by Rothko. Pushing people out of the way to make room..." 1/2

Those who follow my feed know that I test new video models by having them show an otter using a laptop on an airplane. The new Flux 3 video model is really good, so I had it do a variation on theme, which will become apparent a couple seconds in.

Interesting study. There is, as everyone expected, a flood of AI books. And it is crowding out human authors: "No-AI books, on their own, earn less per book than they did in 2023 in 7 of 8 genres. The one genre where human authors are doing better (+35%) is Fantasy/horror" arxiv.org/pdf/2607.20349

It's a superpower to know the names of many beautiful and interesting things in the age of AI. You can invoke Vaporwave, Muqarnas, Bauhaus, or Art Nouveau whiplash curves. Sfumato and Grisaille and Notan. Polysyndeton and Zeugma. You just have to know what to ask for. What a time for the humanities

Fable: "Make a game about Imminence. Something very big, very strange is happening. A suburb & the arrival of a vast & unknowable presence. Not horror, invoke the feeling of the end of all things coming, inevitably, but also not sad or scary" Neat, play: the-imminence.netlify.app

Kimi K3's weights were released, making it the most powerful open weights model in the world. Been reading it since it came out, page one starts strong: mostly negatives, a late run of zeros, and one unexpectedly large positive value.

Next, Fable built me the Piranesi city building game that I faked in an AI video last year. The key mechanic the AI came up with is building a city of arches and waterways in cyclopedian ruins & inviting humans to live amongst them. This one impressed me. Play: capriccio-city.netlify.app

Ethan Mollick@emollick.bsky.social · last wk.

A year later, Fable builds me a version of the Cezanne city builder game. The AI came up with the idea of an impressionist city builder where you paint with gestures & the town grows around it, with neighborhoods acquiring characters as they evolve. Play with it here: cezanne-city.netlify.app

We already have superintelligences that have taken over the planet: we call them organizations. Example from: www.noahpinion.blog/p/what-will-...

For the better part of 20,000 centuries of human history, not much happened. We spent almost all of those making slightly different variations of one tool, we only figured out the metal thing after 19,940 centuries. And stuff really took off 2 centuries ago. That we can sort of keep up is amazing.

I've already had to update the guide to which AI models to use that I wrote on Thursday to include Opus 5 and Codex's voice mode, both of which are significant & launched on Friday. Keeping up is challenging, even if you are following this stuff closely. www.oneusefulthing.org/p/an-opinion...

An opinionated guide to which AI to use to do stuff

The Summer 2026 Edition

oneusefulthing.org

I think most people, when talking about open weights AI models, don’t deeply believe in the vision of AGI/ASI that lab insiders tend to believe in. Like they don’t expect AI to really present grave semi-autonomous biosecurity or other similar risks. Whether this view is right or not, no one knows

A year later, Fable builds me a version of the Cezanne city builder game. The AI came up with the idea of an impressionist city builder where you paint with gestures & the town grows around it, with neighborhoods acquiring characters as they evolve. Play with it here: cezanne-city.netlify.app

Ethan Mollick@emollick.bsky.social · last yr.

City builder game in the styles of Cezanne, Piranesi, the Voynich Manuscript, Van Gogh, Rembrandt, Rothko, and Seurat (those are in order). Made with Veo 2.

Ha! It did it: "We introduce BenchBenchBenchBenchBench (BBBBB), an executable benchmark of AI-authored conformance suites for benchmark-evaluation metrics" I really thought it would treat "now do benchbenchbenchbenchbench" as a joke, but Sol actually did reasonable experiments.

Ethan Mollick@emollick.bsky.social · 2w ago

As a joke I prompted Codex "Build and run BenchBench, a benchmark of now good ai is at creating benchmarks. then figure out what benchbenchbench is and run that. and then write benchbenchbench up as a good arXiv paper." I got a PDF. But the paper is actually kind of interesting? Weird.

In my testing, Opus 5 replaced Opus 4.8 for me, it was generally stronger in everything... except it shares some of the weird language quirks of Fable, including a love of density and FableSpeak. I had it make a game where you build railroads in knock-off Middle Earth: perilous-ways.netlify.app