Martin JJ. Bucher

phd student @Stanford — prev. @Autodesk @StanfordHAI @ETH

[Martin JJ. Bucher @mnbucher]

anything that can be streamlined into tokens will be overtaken by agents and swallowed by the bitter lesson—you either work at the frontier of intelligence, hum2hum vibes, or learn how to bake good bread. anything else seems a suboptimal move given the current status quo.

[tomie @tomieinlove]

(Anthropic): We’re restricting our model to landowners who can pass a literacy test

(OpenAI): Proud to announce our acquisition of Cluely

(xAI): With GrAImes (18+), Grok 6.7 users can experience the thrill of unprotected catgirl coitus without fear of the Woke Mind Virus

[Martin JJ. Bucher @mnbucher]

@BlackHC darkest red flags all over there

[Martin JJ. Bucher @mnbucher]

@evilmathkid still waiting for that update! :)

[Martin JJ. Bucher @mnbucher]

@LombardiHimself vibes are goooood these weeks

[roon @tszzl]

the relative geographic and climate capital of san francisco is at its peak right now. sunny 60 degrees perfect clear skies cute families in picturesque parks while the rest of the country is frozen solid. rich people from the east coast are flying in and filling up the hotels

[Martin JJ. Bucher @mnbucher]

@MattNiessner very nice!! any pointer on how local scene editing is handled? if 3DGS is the output from another, more implicit representation, is editing done in pixel space via full pipeline run?

[Martin JJ. Bucher @mnbucher]

is it? i assumed they just threw A TON of parallel compute with some trained process reward model and then distilled everything into a single final answer for sub? like no dense latent reasoning here it's all still just long streams of tokens i thought:

"Noam revealed that a multi-agent component was used in this IMO project, and it was employed to scale parallel computing. However, he stated that they cannot disclose the specific implementation details."

https://t.co/Qt4D9RLmMM

[Martin JJ. Bucher @mnbucher]

@tszzl @linamkhan if only we knew

[Martin JJ. Bucher @mnbucher]

@SharonYixuanLi while interning full-time 😭

[Martin JJ. Bucher @mnbucher]

@lateinteraction very true unfortunately

[Iro Armeni @ir0armeni]

At @CVPR? Come to the AI Art exhibit to interact with our #GradientCanvas!

It is a community generated #AR painting. Users can claim a region and add their own contribution. No painting skills required, just words!
https://t.co/Cm8R84NxnV

Gradients Spaces, Stanford Univ.

[Martin JJ. Bucher @mnbucher]

@Dopoiro people are working on this direction. definitely lots of potential here.

[Martin JJ. Bucher @mnbucher]

✨What if you could create and redesign scenes by just talking to them?

**ReSpace: Text-Driven 3D Scene Synthesis and Editing with Preference Alignment**

Add, remove, and swap objects simply via natural language, e.g., "add tufted dark gray sofa". (1/8)

[Martin JJ. Bucher @mnbucher]

✨What if you could create and redesign scenes by just talking to them?

**ReSpace: Text-Driven 3D Scene Synthesis and Editing with Preference Alignment**

Add, remove, and swap objects simply via natural language, e.g., "add tufted dark gray sofa". (1/8)

[Martin JJ. Bucher @mnbucher]

Check out our paper!

Work done by myself and the amazing @ir0armeni .

Paper: https://t.co/aoMsX48W8N
Website: https://t.co/p2VBf3ISTM
Code: https://t.co/Qxuok6FLZJ

[Martin JJ. Bucher @mnbucher]

Results: New SOTA on object addition, and competitive on full scene synthesis. Explicit boundaries via SSR significantly reduce out-of-bounds vs. floor plan renderings.

Test-time scaling with Best-of-N shows further potential to improve preference alignment on this task! (7/8)