old, grumpy person in ui-dev-world (aka design-engineering) and AI (yes I am an expert 🫤)
Bremen / Germany
Joined August 2024
- Tweets2.2K
- Following785
- Followers539
- Likes9.1K
Nikolaj Sokolowski retweeted
A few of you asked how this came together. This is the trail of iterations that led to the video!
Who said AI only generates slop?
Second experiment. This time I wanted to see how voice, image and edit could build a climate around Borges’ “Ajedrez”. Fifteen cuts. What do you think?
Stack: Opus 5.5 (x-high), @Paper (editorial), @midjourney.
Yes.
For the next days I will also let claude generate some promo-video stuff it seems - it is an intersting new possibility but not easy to steer.
Here for 3 new pattern on bookofshapes.
Inspired by patterns on turtletoy.net/
(It was pointed out that I may have copied turtletoy.
I didn´t. I did not even now it until then but think it is a great source for anyone interested in generative and geometrical line-art )
From the series: "Can Jev do that?"
Today: a knowledge base (aka second brain).
1.5 years ago, I launched my first larger AI project (Seio, some of you may remember).
I wanted a system where I could drop any kind of content as quickly as possible, from every device I use.
The system then handles sorting and grouping and creates connections between entries (all of this long before Karpathy).
I've now used that dataset (without tags and metadata, of course) to test what something like this would look like with Jev.
What can I say … it works.
The ingest part is obvious: Jev is primarily built to tag, categorize, etc.
But what about retrieval?
Here too, Jev shines, there's no other way to put it.
Without any embeddings and without expensive agentic search, Jev shows me the expected top hits.
Is it perfect? No. But then again, none of the approaches I've tried so far were (and the "perfect approach" probably doesn't exist in this field yet anyway).
For a live implementation, I'd probably still not do without embeddings, and I'd add an LLM here and there.
Used in the right places, though, Jev can significantly reduce retrieval time and noticeably improve answer quality.
Test for the use case "knowledge bases": clearly passed 🎓 ✅
The demo shows Jev-only ingestion AND retrieval.
No embeddings, no LLM.
Nikolaj Sokolowski retweeted
How good is Jev at designing a design system?
You give it one sentence, and it picks from over 1 quadrillion combinations.
Nothing in the video is sped up, each restyle is 4 parallel calls: about 2 seconds, $0.0007
29 flags control corner shape, borders, icon family, chart paths, page texture, density and motion, all on @shadcn. Colour, radius, spacing and type scale are continuous on top of that.
Interesting little Sunday experiment
Nikolaj Sokolowski retweeted
Got access to Jev - not yet sure what use cases I have for it so experimenting for now 😄
Typing a half-remembered movie description into two things at once: Jev on the left, GPT-5.6 Sol on the right, both given the same 258 films.
Every keystroke fires a new call to both. Jev returns a probability for every film in one ~400 ms call, so the wall re-sorts as you type. The chat model has to reread the whole list and write a sentence, so it mostly gets cut off before it finishes.
Jev answered 13 times for 2.4¢. GPT-5.6 got 6 out before being cut off, for 8.8¢ (50¢ at list price, before prompt caching).
JEV ist a monster.
Here: A generative Shop in action.
And this is only one of thousands of use-cases.
Another important building block for the software of the future.
Nikolaj Sokolowski retweeted
New experiment: json-render + jev
The future Generative UI is instant
Your components, your actions, your design system
Rendered in milliseconds
Nikolaj Sokolowski retweeted
Nine dev tools for the details. Part 7.
✦ Cue → cuedesign.space
✦ Bencho → bencho.dev
✦ Evil Buttons → evilbuttons.com
✦ Cursor Lab → cursor-lab-ui.vercel.app
✦ Book of Shapes → bookofshapes.com
✦ Shaders → shaders.com
✦ vgpu → vgpu.sh
✦ Shotbase → shotbase.com
✦ Splitsense → splitsense.ai
New ones every week → productrally.com
For those interested how the heads are constructed and drawn. I let Claude document the whole process (as I usually do with everything experimental) in a nice interactive format.
Does everything make sense? Probably not ...
Is it a good starting point to get some inspiration to start something new? I hope so.
allmyfriendsaremadeofjavascr…
