to help your mental model, usually they don't regenerate it from scratch. lots of simpler apps more or less work out until it needs actual architecture, which the model doesn't realize on its own yet
Great visualization of sampling trajectories in different diffusion formulations
alechelbling.com/blog/rectifi...
who else is competing in the "serves you only relevant work content and respects your time" space?
I also really like dumping relevant state over time + .toMatchInlineSnapshot() but getting AI to use it is hard!
have any tricks?
when you "convert" from a design tool's json to html or vice-versa, a lot of details can shift or get lost in conversion. the only way I could be sure that I would be able to teach the model to perfect the small details would be to have designers design directly in html. there wasn't tool for that!
when trying to teach ai models to design, I needed designers to be able to create the "correct" results where the model could understand it 100% (ie in html + tailwindcss)
prefix is an extension that lets you design in visual canvases within Cursor. you can use Cursor's agents, claude code or any other tool to edit them, because they're just text files!
www.youtube.com/watch?v=QjJe...
want to know what a full "design" mode would be like in cursor? I did too… so I built it!
Screenshot of a design tool showing a chat thread on the left and a generated icon on the right. The chat discusses simplifying an icon, asking for feedback on adding a circle, and notes a 2‑second thought. A label reads ‘Generated image: icon of a man walking a dog, solid bl.’ The icon is a black, minimalist silhouette of a person holding a leash attached to a dog on a white background.
it's pretty cool you can interactively generate images in "antigravity"
Your lettering is amazing! The setup just looks great as a whole.
hard and complicated today is tomorrow's well understood and routine
Fourteen hours later, he's still going.
Thank you, @booker.senate.gov, for using your voice to alert people to Trump and Musk's wrecking ball to Social Security and other vital services Americans rely on.
interesting data structures aren't necessarily printable, also I find that quite annoying haha
I think we will figure out how to teach LLMs to use the debugger and logpoints and then it's all over haha
love this! have you seen people training models that can interpret gestures alongside other modalities like voice?
I messed around but didn't have a great dataset
It's just crazy that we have to defend funding for science in the wealthiest country in history. Censorship, intimidation, and denial of basic truths weaken our future.
I'm a proud mother. Whether I'm walking my son to school or reading him a bedtime story, I do everything to make sure he grows up loved and cared for.
And for the sponsor of the drag ban to say that my existence is a fetish is false & cruel.
I'm glad we were able to defeat this hateful bill today.
the only way the math works out with fixed transforms is adding enough brightness that you're above the all of the background, ie darkening layers or hdr colors.
ios has been using hdr more recently, and of course uses many backdrop layers that measure the colors behind
Good luck with that! Supply is going to be tight.
NAIAD OUT NOW! 🪷
Finally! After 5 years of development and dedication, I'm excited to share my first game with you.
Immerse yourself in a poetic, dreamlike, and serene experience. 😊
✨ Steam (PC/Mac)
✨ PS5 / PS4
✨ Xbox Series / One
✨ Nintendo Switch
