Omer Shapira

@omershapira.com

Graphics ∩ Vision, Electronic Elsewheres. AI for XR @ Meta. Previously : NVIDIA {Research, Omniverse XR, Robotics}, much before that VR, VFX & TV. Thinkpiece Piñata & Artist inbetween. Opinions? No 📍Brooklyn/TLV https://omershapira.com

Google’s presser was so good that I’m begrudgingly looking up XLA tutorials and considering changing my online handle from SIMD Crawford to Crossbarbra Streisand

I was reluctant to take the role of General co-Chair for I3D in a year I was switching jobs and having a child, but people’s willingness to lovingly contribute was something I simply didn’t account for. The program is amazing. I hope you show up. i3dsymposium.org

Huge news: Naty Hoffman will be speaking at I3D 2026. I spent a lot of time brainstorming with him on the appropriate talk to give on the year he retires, and I think what he has is something every researcher will need to hear. Sorry for hyping, but I’m pumped.

I3D Symposium@i3dsymposium.org · 3mo ago

We’re excited to welcome Naty Hoffman as a keynote speaker at I3D 2026. More details: i3dsymposium.org/2026/keynote... Register here: i3dsymposium.org/2026/registr... #I3D2026

Last year I became the caretaker of a 1991 Firebird previously used by Mark Weber of Pulp, during DIFFERENT CLASS, THIS IS HARDCORE and WE LOVE LIFE. I’d like to hang it on a wall, but Firebirds are impossible to hang, so I am forced to play it.

Bild

Graphics / Games students: Sign up to volunteer at @i3dsymposium.org! We're looking for student volunteers for the duration of the conference. Priority will be given to students local to the bay area. DM me or email general@i3dsymposium.org .

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

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.

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.

Sakana AI@sakanaai.bsky.social · 7mo ago

Introducing DroPE: Extending Context by Dropping Positional Embeddings We found embeddings like RoPE aid training but bottleneck long-sequence generalization. Our solution’s simple: treat them as a temporary training scaffold, not a permanent necessity. arxiv.org/abs/2512.12167 pub.sakana.ai/DroPE