Jim Bohnslav @CVPR

post-training VLAs @zoox

[Jim Bohnslav @CVPR @jbohnslav]

Realized I never posted this: I gave a talk at ReInvent this year about our multimodal-language-action model and the infra needed to train it.

link 👇

[Jim Bohnslav @CVPR @jbohnslav]

@vikhyatk image resizing for sure

[Jim Bohnslav @CVPR @jbohnslav]

@code_star I asked 5 times, all got rejected

[Jim Bohnslav @CVPR @jbohnslav]

@YGandelsman @CVPR being finally fundamentally creative is a lot harder than loraing a model on a new task

[Jim Bohnslav @CVPR @jbohnslav]

why the fuck does the escalator at CVPR play a creepy laugh track?

legit thought I was getting schizophrenia

[Jim Bohnslav @CVPR @jbohnslav]

@sleepy_devo the bank won't lend me $500k to put into the sp500

[Jim Bohnslav @CVPR @jbohnslav]

@xeophon don't give away the alpha

[Jim Bohnslav @CVPR @jbohnslav]

@i_ikhatri what do you mean? we've already reorged perception into the alpamayo encoder division, planning into the alpamayo decoder division. sim is alpasim. what else is there?

[Jim Bohnslav @CVPR @jbohnslav]

@jbhuang0604 if you're not training bitter-lesson-pilled world action language models for agentic AI, what are you even doing?

[Jim Bohnslav @CVPR @jbohnslav]

@yoavgo I used to do ML for robotics, but now I do post-training for physical AI

[Jim Bohnslav @CVPR @jbohnslav]

@chris_j_paxton also DINOv3 came out, but it doesn't work as well in the wild as DINOv2. people are trying

[Jim Bohnslav @CVPR @jbohnslav]

if you see me at CVPR

ps @zoox lmk when the robotaxi suit is ready

[Jim Bohnslav @CVPR @jbohnslav]

https://t.co/AX554fzhYL

[Jim Bohnslav @CVPR @jbohnslav]

Realized I never posted this: I gave a talk at ReInvent this year about our multimodal-language-action model and the infra needed to train it.

link 👇

[Jim Bohnslav @CVPR @jbohnslav]

at CVPR in Denver ✈️ If you're interested in (or working on) foundation models for robotics, reasoning VLAs, or adding novel sensors to VLAs, DM me.

#CVPR2026

[Jim Bohnslav @CVPR @jbohnslav]

@yacineMTB many such cases

[Nathan Lambert @natolambert]

My time at Ai2 / @allen_ai has come to an end.

Ai2 is a wonderful place. The last 2.5+ years building Olmo, Tulu, and other projects will be one of the peaks of my entire career. I'm extremely thankful for my teammates and the open community who made this work possible.

For me, it's time to try something different. I will still be working in the open model & open science spaces (more news on that soon). In the meantime I'll be spending a few months learning, chatting with a broader network, getting married (!!) and most importantly recharging from pouring my soul into this place.

I've attached the note I shared with the team and some fun photos from our time together. I'll keep cheering for Ai2 and am excited to see what you build next.

[Zhipeng Cai @cai_zhipeng]

🎇Thrilled to release VLM^3! Most 3D vision papers nowadays still spend months/years designing complex archs/losses/augmentations for different tasks. Are they necessary? VLM^3 shows that most designs that you think are important for 3D vision are [not] important at all!

[Jim Bohnslav @CVPR @jbohnslav]

@hq_fang couldn't find that one 🤔

[Jim Bohnslav @CVPR @jbohnslav]

@simonkalouche @KyleVedder @astridwilde1 if you figure they subsidized the hell out of FSD development for a long time, that's sort of paying for data