Veo 2 prompt: "a distant shot zooms in to reveal a knight wearing a golden helmet, he begins to charge on his zebra, lowering his lance, charging towards a clockwork octopus" (this is one of the initial 4 videos it made)
"an woman with short black hair assembles an impossibly complicated device, close up on her face, she is sweating"
The consistency of small details is really impressive, the fact that the shaft of the screw turns at the same speed and direction, hair and sweat, tattoos...
"a man holding the leash of a golden retriever stares mournfully at a fireworks display over his small town of Tudor-style homes, the flashes punctuate the darkness."
All videos are from the first 4 from the prompt. I did learn that you can't ask for many cuts or scene changes.
"a low angle pov drone shot flies along a beach as a vast beast ascends from the water"
Water dynamics are pretty impressive
"A commander of a space ship, wearing a futuristic military uniform with a mechanical parrot on her shoulder; walks around giving orders on a damaged and burning starship bridge"
So here is a failed case. Prompting for a complex gymnastics routine at a crumbling castle, you can see that model still can't do what is the hardest AI video edge case right now - gymnastics. Tumbling + multiple occluded limbs + human anatomy pushes these models to their limits.
"black and white documentary footage of the filming of the Apollo Moon Landing on a soundstage, with appropriate 1960s era equipment"
I think the lighting on the woman's face is super interesting here, as is the physicality of the interaction with the equipment.
I continually get great initial results from a single prompt. It feels crazy to be able to summon these videos into existence.
"a woman stands on featureless high tower made of concrete with veins of gold and lapis, below waves crash as a stormy ocean spreads in all directions"
"a pug is crowned king of england in an elaborate ceremony, ending with a crown being placed on its head"
I liked all of the initial videos so much.
"The camera swoops through a hyperealistic space evocative of piranesi and escher, filled with stairs and ruins, a couple holding hands walks down one of the stairs"
Impressed by the shadows, the solidity of the people's feet on the stairs, the switching of the held hands...
• • •
Missing some Tweet in this thread? You can try to
force a refresh
ChatGPT 5.2: "Build an interactive Excel spreadsheet where I can pick two D&D monsters to fight against each other and the spreadsheet simulates the combat somehow, including special abilities. Give a D&D look"
Thinking took 60 minutes(!) & had to have it fix an error, but cool
Claude 4.5 Opus followed the same instructions very quickly, and with style, but simplified the problem to avoid using actual special abilities or status, just straight up rolls for damage
Gemini 3 Pro. I really hope they add consistent ability to work with or download files.
I did not expect that the PowerPoint killer would be something called Nano Banana Pro, but that is where its heading
It makes the major efforts by all the other AI companies, including Microsoft, to crack PowerPoint by using python seem like a dead end
ImageGen is all you need?
The thing is that NotebookLM can just take source materials, a topic, and an idea and make a very pretty, impactful deck.
Hallucinations are very rare, though there are still some spelling and graphics issues. Editing capability is apparently coming, but the direction is clear.
The slide deck is the result of me throwing my entire book into NotebookLM, by the way.
Voice is one of the most useful ways to interact with AI to do work but it seems to have been semi-abandoned for serious use outside of the “chat with a friend” case.
All of the voice modes only access weak models with low latency, making them zippy & fun but kind of useless.
If you don’t think of voice models as a fun chat, but rather as a way of working, it suggests that pauses are fine, even preferred (don’t talk with me unless you have something to say). And alternative UXs beyond “talk with your AI about the weather” become possible to explore.
Also I want to turn off the breathing, giggling, and disfluencies. Anthropomorphism can be helpful in many cases but it gets to be too much, especially for serious discussions. The tone is off and it feels ingratiating and slows things down.
I think my “otters on a plane using WiFi” may be a saturated benchmark now that nano banana pro can do this.
Prompt: Scientists who are otters are using a white board to explain ethan mollicks otter on a plane using WiFi test of AI (you must search for this) and demonstrating it has been passed with a wall full of photos of otters on planes using laptops
Since there are so many AI announcements, my advice is to focus on those expanding what folks can do with AI (& especially tools that democratize who can use AI) rather than every single UX improvement
Skills, connectors & agents with file access/CLIs are especially interesting.
Next up: pay attention to expansions in artifacts/vibe coding for non-coders, specialized AI tools for industries outside of coding (see Claude Finance) and systems that take software people use every day and radically transform how they work using AI (Excel agents, for example)
Also interesting to watch ambitious new applications that are AI-native. What Google is doing with NotebookLM, for example, is basically creating an entirely new interface for working with information that is a pretty strong break with older ways of handling large amounts of info
I don’t have much to add to the bubble discussion, but the “this time is different” argument is, in part, based on the sincere belief of many at the AI labs that there is a race to superintelligence & the winner gets,.. everything.
You don’t have to believe it (or think this is a good idea), but many of the AI insiders really do. Their public statements are not much different than their private ones.
Without considering that zero sum dimension, a lot of what is happening in the space makes less sense.
This is not the only way folks justify the large spend on AI buildout (and whether there is a bubble seems very far from obvious), but it is a dimension that does not show up in as many economic analyses as it should.