Ilya Zisman

75 posts

Ilya Zisman banner
Ilya Zisman

Ilya Zisman

@suessmannn

RL i guess | https://t.co/AFHWgRKvrt | PhDing

Universal Traveler Katılım Ekim 2021
594 Takip Edilen346 Takipçiler
Sabitlenmiş Tweet
Ilya Zisman
Ilya Zisman@suessmannn·
🔥 Zero-shot generalization is the dream: adapt instantly, no fine-tuning. It's why LLMs blew up—but it's not just a language modeling thing. It’s happening in RL too. 🚨 @maxsbob21's new paper dives deep into zero-shot RL under shifting dynamics—and why current methods break.
Ilya Zisman tweet media
English
4
20
143
14.3K
Alexander Nikulin
Alexander Nikulin@how_uhh·
This was a long journey, and its not over yet!! thanks for cringy photo, Ilya
Ilya Zisman@suessmannn

My dearest friend Sasha @how_uhh, congrats on successfully defending your AI PhD! Everyone please address him as Dr. Nikulin in the comments from now onwards! ps: who’d have thought during our freshman year as Sociology majors [pic taken ~10 yrs ago] that I'd posting this tweet?

English
4
0
14
591
Ilya Zisman
Ilya Zisman@suessmannn·
My dearest friend Sasha @how_uhh, congrats on successfully defending your AI PhD! Everyone please address him as Dr. Nikulin in the comments from now onwards! ps: who’d have thought during our freshman year as Sociology majors [pic taken ~10 yrs ago] that I'd posting this tweet?
Ilya Zisman tweet media
English
0
0
17
861
Ilya Zisman retweetledi
Nikita Kachaev
Nikita Kachaev@judokach·
👀 Action fine-tuning often blinds VLA models: they lose the visual–language (VL) priors that made them smart. We show how to keep those priors intact with a tiny alignment loss. 🤖 ↓
English
5
20
110
7.5K
Ilya Zisman retweetledi
Denis Tarasov
Denis Tarasov@ML_is_overhyped·
I’m asking for help. I was meant to start my PhD with @_rockt and @robertarail at UCL, but my UK background check was refused. My appeal seems unlikely to succeed, so I’m urgently searching for any PhD or research positions in academia or industry. Any help is appreciated.
English
14
35
251
40.8K
Ilya Zisman retweetledi
vik
vik@vikhyatk·
@GFaang97609 no i never run ablations, because i believe god has a plan
English
8
19
192
14.7K
LeRobot
LeRobot@LeRobotHF·
LeRobot SO101 setup just got 50% cheaper! You can now teleoperate your follower arm right from your phone. 🤯 But that's not all. Our new pipeline feature lets you record and train AI models in end-effector space, or with any other features. The possibilities are endless!
English
11
65
444
54.7K
Ilya Zisman retweetledi
Alicja Ziarko
Alicja Ziarko@ZiarkoAlicja·
Can complex reasoning emerge directly from learned representations? In our new work, we study representations that capture both perceptual and temporal structure, enabling agents to reason without explicit planning. princeton-rl.github.io/CRTR/
English
6
119
856
89.6K
Michał Bortkiewicz
Michał Bortkiewicz@m_bortkiewicz·
🚨New paper How effectively can search be captured in representations? Can the combinatorial structure of problems (Rubik/Sokoban) be embedded in these representations? These and other questions we tackle in the new work led by @ZiarkoAlicja. Check it out!
Alicja Ziarko@ZiarkoAlicja

Can complex reasoning emerge directly from learned representations? In our new work, we study representations that capture both perceptual and temporal structure, enabling agents to reason without explicit planning. princeton-rl.github.io/CRTR/

English
1
1
8
287
Ilya Zisman
Ilya Zisman@suessmannn·
@aaronwetzler wow that's something I see for the first time on twitter with such money/value tradeoff
English
1
0
2
105
Aaron Wetzler
Aaron Wetzler@aaronwetzler·
Enabling useful teleoperation for complex dexterous robotic arm tasks in decent working spaces proved challenging with the open source SO100 design ($100 arms). Reach, degrees of freedom, and strength were insufficient. The next price point ($1000) felt unnecessary. We’ve been customizing low-cost arms ($300) for useful dexterous manipulation. Despite some challenges the teleop results are now really promising. Maybe $200-300 is the sweet spot. Curious about others’ experiences.
English
5
2
24
964
Gabriele Berton
Gabriele Berton@gabriberton·
SLAM is dying? I had a chat with roboticists from Stanford and Tesla not long ago and (to my surprise) their consensus was that SLAM is dying. For 2 reasons: 1) robotics is moving towards end to end in many fields (see @comma_ai openpilot, Tesla, @physical_int). These (1/4)
Gabriele Berton@gabriberton

My post sends the wrong message. Here's a clarification + some thoughts (1) SLAM has been solved for ages (without deep learning) (2) IRW deep learning is bad for SLAM (3) SLAM is dying (4) VGGT and ORB-SLAM are great, they just focus on different tasks Let me elaborate (1/5)

English
8
7
136
64.1K
Ilya Zisman retweetledi
Jay Yang
Jay Yang@Jayyanginspires·
As boring as it sounds, I’m slowly realizing that 90% of success is doing the obvious thing for a painfully long amount of time without convincing yourself you’re smarter than you are.
English
158
2.1K
20.7K
556.3K
Ilya Zisman
Ilya Zisman@suessmannn·
Had a blast together with @how_uhh at @LeRobotHF hackathon this weekend. Built phone-based teleoperation for my SO-100 arm using pose estimation. Here’s a quick BTS of the final demo with teleop working (+ a small victory dance 🎉)
English
26
42
411
54.7K
AlexC
AlexC@Beeg_Brain·
@suessmannn @LeRobotHF Thank you, all the arms are original SO-100 except one that is the SO-101 (the blue one on the left) :)
English
1
0
1
69
AlexC
AlexC@Beeg_Brain·
If you missed it yesterday, I released the full demo video of our robot drawing pixel art from @LeRobotHF worldwide Hackathon (thread below) 👇I would be happy to discuss in detail about it for interested people🤗(blogpost coming soon on HF)
AlexC@Beeg_Brain

🚀 (1/3) Built a LEGO pixel art robot in just 48h at the @LeRobotHF hackathon by @huggingface! 🎨 You draw a 5x5 pixel design → the system builds it with LEGO bricks using multi-arm coordination + modified ACT policies. 👇 Watch the robot in action: youtu.be/GeEeOWr0rMs

English
1
5
65
4.8K