Roei Herzig (@roeiherzig) / X

Roei Herzig

2,336 posts

Roei Herzig

@roeiherzig

Postdoc @berkeley_ai. Researcher @IBMResearch. PhD @TelAvivUni. Working on Multimodal Robot Foundation Models and Structured Physical AI.

Berkeley, CA

Joined March 2017

Replies Articles Media

Pinned
Roei Herzig
@roeiherzig
Jun 18
If you want a vision encoder for dexterous manipulation, what should be the most important part to model? 🤔 Current standard models like CLIP, SigLIP, and DINOv2 have an incredible grasp of semantics and spatial details. But they lack the action-centric structure needed for
00:00
Roei Herzig
@roeiherzig
Jun 6
Article
CAIP: Contrastive Action-Image Pre-training for Visuomotor
This is a (short!) story about how we developed CAIP, an action-centric encoder for dexterous manipulation. If you would like to have a vision encoder for dexterous manipulation (or robotics), what...
Roei Herzig
@roeiherzig
12h
How could there be a consensus if nobody knows what their model even *is*? 😂
Varun Nair
@_varunnair
Aug 6
Is there consensus that @GeneralistAI the best public physical AI model right now? Who else is close? @physical_int
Roei Herzig
@roeiherzig
Aug 6
Another Sunday in Mountain View, yet another tweet that was completely written by Claude, that may or may not have ever happened in real life.
Ravid Shwartz Ziv
@ziv_ravid
Aug 5
A sunny day in Mountain View. Like every other day. Jeff, Sanjay, Oriol and Quoc walk into a VC office on Sand Hill. The VC is on his second oat latte and has fourteen minutes blocked for this. Jeff: So. We decided to build our own company. VC: Okay, great, love that. Who's
Roei Herzig
@roeiherzig
Aug 4
וואי, גמיני כזה מודל פח, אבל סול 5.6 מדהים. לגמרי שיחקו אותה oai.
Boaz
@boazbe
Aug 3
משתמש עכשיו בClaude Code לפרויקט לא קשור לעבודה והולי פאק כמה המוצר Overhyped ולא באמת frontier. נראה לי שכל ההתלהבות זה כי כולם משתמשים בFable על המקסימום Thinking ועובדים על מערכת CRUD גנרית, כי Opus עם High לא מדלבר דברים רציניים, איטי בטירוף, ויוצר slop גרוע שחשבתי שכבר עבר
Roei Herzig
@roeiherzig
Aug 3
It all comes down to how you write the paper. If the paper positioned itself as 'while previous work shows X, we show how scaling can be achieved in diffusion', then sure. But if your main claim is 'We discovered a third pre-training axis beyond parameters and data:
Aadim Nepal
@NepalAadim
Aug 3
I have read all discussion points and I must say Alexi is 100% correct here. To people who haven't read this work properly: stop all this non sense theatrics. Take a time to read this work properly, and only then you will appreciate it. Alexi is the first person to show this