Chaim Baskin (@ChaimBaskin) header
Chaim Baskin (@ChaimBaskin) profile picture

Assistant Professor @bengurionu | Fundamental Research Lead @DecartAI | Building efficient, robust foundation models, world models & real-time generative AI

Israel
Joined November 2019
Photo by Chaim Baskin (@ChaimBaskin)Photo by Chaim Baskin (@ChaimBaskin)
1/7 Some of science's greatest missed opportunities aren't failed experiments. They're discoveries that never get the chance to change the world. Yesterday, at Nevo Labs and Yad Hanadiv's Bridge to Impact event in Jerusalem, I shared why that matters. ๐Ÿงต
1
2
10
90,041
6/7 That's the bridge at its best: practical challenges reveal new scientific questions, while discoveries open new possibilities for real systems. It was inspiring to see Nevo Labs and Yad Hanadiv actively supporting this vision through the Bridge to Impact Grant.
1
81
7/7 Grateful to Scott Tobin for moderating, to fellow panelists Lihi Adler-Abramovich and @RosenblumSerge Serge Rosenblum, and to everyone who made the evening possible. We should be as ambitious about what discoveries can become as we are about making them.
76
6/7 That's the bridge at its best: practical challenges reveal new scientific questions, while discoveries open new possibilities for real systems. It was inspiring to see Nevo Labs and Yad Hanadiv actively supporting this vision through the Bridge to Impact Grant.
1
32
7/7 Grateful to Scott Tobin for moderating, to fellow panelists Lihi Adler-Abramovich and @RosenblumSerge , and to everyone who made the evening possible. We should be as ambitious about what discoveries can become as we are about making them.
23
Chaim Baskin retweeted
ืžื•ื“ืœ ื•ื™ื“ืื• ื‘ื–ืžืŸ ืืžืช ื—ื“ืฉ, ืžืจืฉื™ื ื•ื–ื•ืœ - ืจืง $0.02 ืœืฉื ื™ื”. ืžื•ื“ืœ Lucy 2.6 ืฉืœ Decart ืžืืคืฉืจ ืœืขืจื•ืš ื•ื™ื“ืื• ื‘ื–ืžืŸ ืืžืช ื‘ืืžืฆืขื•ืช ื˜ืงืกื˜, ืชืžื•ื ืช ืจืคืจื ืก ืื• ืฉื™ืœื•ื‘ ืฉืœื”ื. ื‘ืขื‘ืจ ื”ื™ื™ืชื™ ื‘ื•ื ื” ืชื”ืœื™ื›ื™ื ื“ื•ืžื™ื ืขื ืžื•ื“ืœื™ SAM ืœืกื’ืžื ื˜ืฆื™ื” ื•ืžืขืงื‘ ืื—ืจื™ ืื•ื‘ื™ื™ืงื˜ื™ื, ื•ืžืฉืœื‘ ืื•ืชื ืขื ื›ืœื™ื ื ื•ืกืคื™ื ืœืขืจื™ื›ื” ื•ืœื”ืจื›ื‘ื” ืฉืœ ื”ืชื•ืฆืื”. ืžืœื ืขื‘ื•ื“ื” ื•ื”ืชื•ืฆืื” ื”ื™ื™ืชื” ืกื‘ื™ืจื” ืœื›ืœ ื”ื™ื•ืชืจ. ื›ืืŸ ืžื–ื™ื ื™ื ืกื˜ืจื™ื ืฉืœ ื•ื™ื“ืื• ื•ืžืงื‘ืœื™ื ืกื˜ืจื™ื ืขืจื•ืš ื‘ื—ื–ืจื”. ื”ื—ื™ื‘ื•ืจ ื ืขืฉื” ื“ืจืš WebRTC, ืขื ืชืžื™ื›ื” ื‘-720P ื•ื‘-30 ืคืจื™ื™ืžื™ื ืœืฉื ื™ื™ื”. ืืคืฉืจ ืœืฉื ื•ืช ืืช ื”ื”ื ื—ื™ื•ืช ืชื•ืš ื›ื“ื™ ื”ืขื‘ื•ื“ื”. ืœืคื™ ืกืจื˜ื•ื ื™ ื”ื”ื“ื’ืžื” ื•ื›ืžื” ื“ื•ื’ืžืื•ืช ืฉืจืื™ืชื™ ื”ืชื•ืฆืื•ืช ื ืจืื•ืช ืžืขื•ืœื•ืช. ืื ื™ ื›ืžื•ื‘ืŸ ืื ืกื” ื‘ืขืฆืžื™. ื–ื” ื›ืœื™ ื—ื–ืง ืžืื•ื“ ืœื”ืคืงื•ืช, ืœืจืื•ืช ืชื•ืฆืื•ืช ื‘ื–ืžืŸ ืืžืช ื•ืœื ื‘ืคื•ืกื˜ ืคืจื•ื“ืงืฉืŸ ืฉืœื•ืงื— ืฉื‘ื•ืขื•ืช. ื”ื‘ืขื™ื” ื”ื™ื ื›ืžื•ื‘ืŸ ืื‘ื˜ื—ื” ื•ืขื•ื“ ืฉืœืœ ืืคืฉืจื•ื™ื•ืช ืฉืœ ืžืชืงืคื•ืช ืกื™ื™ื‘ืจ ืžืชื•ื—ื›ืžื•ืช. ืขื•ื“ ื›ืžื” ื—ื•ื“ืฉื™ื ื›ื‘ืจ ื™ื”ื™ื” ื‘ืœืชื™ ืืคืฉืจื™ ืœื”ื‘ื“ื™ืœ. ืขื›ืฉื™ื• ื’ื ื‘ื–ืžืŸ ืืžืช. ืœืคื™ Decart, ื”ืฉื™ืคื•ืจ ื”ืžืจื›ื–ื™ ื‘ื’ืจืกื” ื”ื—ื“ืฉื” ื”ื•ื ื”ื ืืžื ื•ืช ืœืจืคืจื ืก: ืคืจื˜ื™ ืคื ื™ื, ื‘ื’ื“ื™ื ื•ืžื•ืฆืจื™ื ืฉื ืฉืžืจื™ื ื˜ื•ื‘ ื™ื•ืชืจ ืœืื•ืจืš ื”ืชื ื•ืขื”. ื”-API ื–ืžื™ืŸ ื›ื‘ืจ ื‘ืžื’ื•ื•ืŸ ืคืœื˜ืคื•ืจืžื•ืช. ืขื›ืฉื™ื• ื ื›ื™ืŸ ื–ืจื™ื– ืืคืœื™ืงืฆื™ื” ื•ื ืจืื” ืžื” ืืคืฉืจ ืœืขืฉื•ืช ืื™ืชื•.
3
2
41
3,474
Chaim Baskin retweeted
Introducing Lucy 2.6. The world's leading live video model, now with a major quality upgrade. Hereโ€™s whatโ€™s new ๐Ÿงต
50
55
18
557
30,357
Chaim Baskin retweeted
Over the past few days, I dug into the geometry of LoRA to understand how exactly it constraints the geometry of the search space. There's a beautiful connection to the *intrinsic dimension* of a task. Also learnt I'm saying "difficulty" wrong... ๐Ÿคก 0:00 MNIST vs the walking humanoid 1:05 Do we still need fine-tuning? 3:05 What is LoRA? 6:55 What does the rank constraint actually do? (geometric intuition) 9:50 Intrinsic dimension: training along random directions 13:30 Intrinsic dimension of pre-trained LLMs 15:10 Demo: fine-tuning a query router with Qwen3.8-27B
14
23
7
333
21,422
Chaim Baskin retweeted
Your first glimpse of Lucy 2.6 has a spooky side... Turn on your camera and see what looks back. Dare to look? halloween.decart.ai
26
8
2
61
5,119
Chaim Baskin retweeted
Our spatial memory system helps video world models reconstruct familiar places. 94x less storage. 44ms retrieval from a 1.65-hour archive. Learn more โ†“ research.decart.ai/spatial-mโ€ฆ
1
2
40
4,313
Chaim Baskin retweeted
Students sometimes ask me if it still makes sense, in this accelerating age, to pursue a PhD in AI. Perhaps counterintuitively, I think it's a great time to do so. I wrote up some thoughts on this here: web.mit.edu/phillipi/www/wriโ€ฆ
90
580
109
3,853
1,747,389
Chaim Baskin retweeted
What if an AI model could take back a harmful answer before it finishes it? Diffusion language models can. New Decart research accepted to #EMNLP2026 found harmful responses in just 9โ€“21% of attacks, vs. 48โ€“62% for comparable autoregressive models. We turned that behavior into SRI Guard - detecting harmful generations with just 0.01โ€“0.04% added latency. Safety shouldnโ€™t come at the cost of speed. Read the paper: research.decart.ai/step-wiseโ€ฆ @INSIGHTLabBGU @TechnionLive @bengurionu @CS_Uni_of_Haifa
3
4
2
52
12,970