Excited to announce our latest (submitted to) SIGBOVIK 2026 @harryqbovik.bsky.social paper: "SchmidhubAI: Accurate Historical Paper Attribution". We built an AI system that, given any modern AI paper, automatically determines which of its ideas were already published by Jürgen Schmidhuber.
the above is still true for python DSLs, but we managed to have Pytorch JIT and Warp be effective subsets with enough context. But I’d love an example. Not that it matters. It doesn’t.
Not that it matters, but I’m now curious
I don’t even mean that extremely. but the cost of maintaining a new compiler and integrating a new language with new edge cases, while managing a sweeping refactor for code that’s entirely out of distribution, is a lot of engineering anti-patterns to fight given the new landscape
You have to give Tim credit for being a pragmatist too. A lot has changed since December
I can’t share the source, but multiple people with firsthand knowledge confirmed it.
So fast that I get to stand at the event horizon and call my buddies on either side on a landline
I’m sad to have to push myself away from thinking about programming tools. It’s a subject I love, have built immensely on top of, but no longer an abstraction worth pursuing, unless building hardware
1. Sad that the Verse programming language is cancelled. It had some brilliant ideas that got tainted by only being available in lisp-likes
2. Beyond sentimental value, it doesn’t matter. Programming languages don’t matter anymore
sed -i '' 's/❎ Have Child/✅ Have Child/g' TODO.txt
The Registration for I3D 2026 is now open! Early bird prices are available until 17th of April, so make sure to register by then for the best price!
View the full details at: i3dsymposium.org/2026/registr...
#I3D2026
also when this is RIGHT THERE
Being interdisciplinary mostly means everyone agrees your work is interesting but not for their journal.
Chairing a conference is basically a flex of soft skills, what with writing soliciting emails, fitting a budget, and enforcing deadlines.
No wonder scientists hate it.
The coming @i3dsymposium.org will take place this May on the Lucasfilm campus in SF.
I will be serving as General Co-Chair.
We rely on industry collaborations and sponsorships to keep the quality high.
If you’re in a position to make I3D better, email: info@omershapira.com.
i3dsymposium.org/2026/
One of my favorite findings: Positional embeddings are just training wheels. They help convergence but hurt long-context generalization.
We found that if you simply delete them after pretraining and recalibrate for <1% of the original budget, you unlock massive context windows. Smarter, not harder.
C++ accomplished that around 1994
Here’s my playlist of 50 bangers (not anthems) from 2025. Have a listen, maybe you’ll hear something you missed. Happy new year.
Spotify:
open.spotify.com/playlist/3xy...
who decided to call it "computational social science" and not "virtue signal processing"
#SIGGRAPH2025 Advances in Real-Time Rendering in Games course talk recording of "FAST AS HELL: IDTECH8 GLOBAL ILLUMINATION" by @idsoftwaretiago.bsky.social from id Software is now online: youtu.be/VTrdeqMMMK0?... Enjoy!
NEWS.
I recently moved to Meta to work on AI for headsets and input systems.
My focus since the beginning of my Engineering career has been “Intelligence Close to the Senses”, and this feels like a natural step in the same direction.
I’ll still be active in AI and Graphics conferences.
if they’re not blocking JIRA as well, they’re not solving the healthcare problem
Nothing like spending hundreds of hours building a guitar from wood and wire to make you appreciate that you can purchase a precision-machined electric instrument with wood from 3 continents and rare earths from a fourth on Black Friday for the cost of 2 family meals
from Gabe Gear to GabeCube
4 layers of black stain, 4 layers of various walnut stains, 4 layers of tung oil and 4 layers of wax later: I’m really really scared to drill the final pieces into place.
Headed to Tel Aviv, Hong Kong, Taipei and Seoul this month. Come say hi!
y'ever see my blog post about Max Payne 2's character shadows? got some good Remedy dev scoops in there
www.joewintergreen.com/dev-scoops-m...
If you have a UV map view superimposed with the faces, you can do a lot of early error detection. far more if you can visualize double lines and normals.
