Hsuan Su (@jacksukk) header
Hsuan Su (@jacksukk) profile picture

Ph.D. student @ntu_spml Research on Data Synthesis

Taipei
Joined October 2012
Photo by Hsuan Su (@jacksukk)Photo by Hsuan Su (@jacksukk)Photo by Hsuan Su (@jacksukk)
Hsuan Su retweeted
The RL era of robotics is here, and simulation is how to scale it. An early look at our research on a achieving a better scaling paradigm for robot learning with simulation 👇
93
164
104
1,591
480,805
Hsuan Su retweeted
Locus sets the frontier for automating molecular design Our automated research system Locus develops scientific AI models that outperform SOTA expert solutions Unlocking the potential of scientific AI requires leveraging proprietary data. Locus achieves this autonomously 🧵👇 -- We test Locus across a number of tasks related to molecular scoring and design: training molecular property prediction models for a proprietary task, fine-tuning Boltz-2 to predict small-molecule potency, and building antibody design workflows around BoltzGen, Boltz-2, ESMC, and ESMFold2. Each of these settings are real-world AI for science use cases, where tasks are often out-of-distribution for existing foundation models and limited in data, while featuring expensive and noisy evaluators. Compared to state-of-the-art baselines and LLM agents set to optimize these tasks, Locus achieves higher performance. Additionally, Locus generalizes better despite limited data, finds more sample efficient solutions, and explores a higher diversity of solutions.
30
38
28
166
54,923
From the early days to the frontiers of mathematics. It’s been so cool watching Carina and the Axiom team build toward moments like this. What a milestone!
The largest math proof in human history is now computer-verified. Today, Axiom announces a complete Lean auto-formalization of the Classification of Finite Simple Groups, the "periodic table of symmetry." About 15,000 pages, built by over 100 mathematicians across generations.
2
4
33
3,661
Introducing a routing control framework to train agentic MoEs: aligning turn-level experts with agent actions while encouraging token-level consistency leads to major performance gains. arxiv.org/pdf/2610.07332 @lblaoke Ting-Yao Hu @cydhsieh @schowdhury671 @OncelTuzel
6
12
457
Hsuan Su retweeted
Hosting a happy hour for people building in robotics, industrial automation, and simulation tomorrow. Come join us for a relaxed evening at one of our favorite speakeasies in San Francisco. We’re keeping it intimate, so space is limited! Apply here: luma.com/kl6l3ry6
15
5
59
7,572
Hsuan Su retweeted
Been seeing a lot of Astra for real2sim, but what you need for robotics: physics and affordance. Our SOTA recipe specifically solves for: • Affordance: model object details that convey functionality. • Physics: calibrate simulation dynamics against real-world observations. Our agent has the action space of a “full-stack simulation engineer” and jointly considers geometry, materials, colliders, articulation, and physics. It automatically calibrates physics against real trajectories and matches task-specific dynamics. Simulation environments is useful only if it can provide meaningful policy level signals. Methods, comparisons, and results below.
31
66
15
585
125,734
Hsuan Su retweeted
The models are improving the models. Locus, our automated AI research system, is SOTA on PostTrainBench and post-trains Qwen3 base models that surpass the human post-trained Qwen3 model. Today, LLMs post-trained end-to-end by Locus are in production to millions. 🧵👇 PostTrainBench evaluates agents' ability to post-train models on various domains given 10 H100 hours. We extend PostTrainBench via PostTrainBench+, which has a greatly expanded compute budget that provides clearer signal on automated post-training capabilities. We find that thousands of H100 hours help distinguish methods' performance post-training Qwen3 1.7B-Base models, and that Locus scales best. In this setting, modes trained by Locus collectively surpass the perforamce of the offical human post-trained Qwen3 1.7B model. In a test of generalization, we ran Locus on all live Kaggle competitions with prize money and public leaderboards. After 16 days, Locus achieved the 4th highest average rank among all participants.
15
59
28
308
270,721
Hsuan Su retweeted
This Thursday in SF. Come by to say hi, will be sharing some of our latest! luma.com/lk9se5mi
Today, we're announcing not one, but two of the @moonlake team on stage at AUTONOMOUS. @sharonal_lee and @sunfanyun, Co-Founders of Moonlake AI, will be on our Frontier Stage to discuss Where Robots Learn Before the Real World. 🗓️San Francisco, July 16th 📍 The Midway 🎟️ autonomousfuture.co Moonlake builds generative world models that turn a plain-language prompt into an interactive 3D environment in minutes. From text to the system compiling a training experience with physics, agents, and logic. Out of @Stanford's AI lab, and then again out of stealth in late 2025 with a flying start, with a round led by NVIDIA's NVentures & others.
1
3
21
2,849
Hsuan Su retweeted
Replying to @chrmanning
@chrmanning is sharing our approach to embodied general intelligence at AI Engineers! Room 2006
6
1
29
12,350
3 years ago I was a researcher at @AIatMeta frustrated with LLM evals. Today we are announcing our $50M Series B to develop agent simulation research and infrastructure 🚀🚀🚀 Also releasing a preview of our Digital World Model 👀
Today, we’re excited to announce our $50M Series B, led by @GreenfieldVC (formerly TPG Capital), with participation from @lightspeed and @notablecap. 🚀 At @PatronusAI, we develop simulations and evals to train and improve AI. The first phase of AI was built on static benchmarks, but that era is over now. As agents are used to solve longer and longer tasks, they need to practice in dynamic, living worlds to get better. Simulations are the critical infrastructure powering this next phase. As a company, we’re behind the most influential research and products in AI evaluation, like FinanceBench, Lynx, and Percival. And things have moved at the speed of light since. ⚡ We partner with the world's leading frontier AI labs and enterprises, and our revenue has grown more than 15x over the past year. Additionally, today, we’re introducing a preview of the first Digital World Model for AI agent training and simulation: Patronus-DWM. Digital World Models are language diffusion world models that predict realistic environment behaviors and steer agent actions across digital workflows. Just as physical world models predict how objects move through space, we’re developing the equivalent for the digital world: predicting how agents act in digital workflows, then using that to scale the creation of high-quality training data for LLMs. Digital World Models help us push the frontier of ultra long horizon workflows, and unlock a new class of self-improving RL environments. This is our scalable approach to simulating all of the world’s intelligence. The round was also joined by @datadoghq, @SamsungVentures, @gokulr, @factorialcap, and a large cohort of amazing AI leaders and researchers across @AnthropicAI, @OpenAI, @GoogleDeepMind, @nvidia, @Recursive_SI, and more. ✨ It has been the ride of a lifetime. But we’re just getting started. The best is yet to come. "Do not go gentle into that good night, Rage, rage against the dying of the light" - Dylan Thomas (1954)
2
4
16
1,370
Hsuan Su retweeted
🚀
Taking main stage at AWE 2026 to demystify world models alongside @bilawalsidhu. 9:55–10:45 AM, Grand Ballroom. Details in comments #awe2026
3
12
2,398
Taking main stage at AWE 2026 to demystify world models alongside @bilawalsidhu. 9:55–10:45 AM, Grand Ballroom. Details in comments #awe2026
4
5
2
53
10,760
Hsuan Su retweeted
Thank you to everyone who came out to Moonlake’s Happy Hour at CVPR! We had 750+ sign-ups overall, so we couldn’t accommodate everyone this time - but our team will be around for the next few hours if you’d still like to chat. Grateful for the energy and support around Moonlake!
2
3
62
15,419
Hsuan Su retweeted
Join us tomorrow at CVPR for great food and even better company! We have an amazing group of researchers, founders, and industry leaders attending. We’re currently over capacity, but if you’d like to join, DM me a short blurb about your research and we’ll do our best to fit you in. luma.com/fxf4l3hv
3
6
33
9,832
Hsuan Su retweeted
Our team at Apple MLR is presenting 4 papers at #CVPR2026! If you're attending, stop by and chat with the authors about our research. 🧵 📌 Velox — Malik et al. apple.github.io/ml-velox/ Jun 6, 11:45 AM MDT 📌 AMusE — Chowdhury et al. arxiv.org/abs/2512.16250 Jun 6, 4:45 PM MDT
1
6
32
1,879
RT @VioletNPeng: My first paper at Google is out! Thank you @rohanpaul_ai for highlighting LEAP. To share more thoughts on this direction:…
This quoted post is unavailable.
1
6
Hsuan Su retweeted
Heading to CVPR June 4–7. Shoot me a DM if you’d like to chat or have recommendations for events, talks, or meetups! I’ll also be speaking about Moonlake and our journey so far: luma.com/nc556ega Come and say hi if you’re around!
3
1
34
63,058
Warehouse example.
1
2
10
1,911
We generate both the engine code and the reskinned, photorealistic outputs. DMs are open if either sounds relevant to you! :)
Real-time neural rendering model @moonlake Works on from any physics/game engine to close the sim-to-real gap. If this is relevant to your team, I'd love to chat (my DM is open)!
5
8
86
27,581
Here’s a sneak peak into a fully generated articulated, sim-ready engine output based on only input image. We work with point cloud and video inputs as well. Any annotations needed i.e., segmentation comes for free as we have full visibility and control all of the engine code base. Medical room example. Reference image (left), 3D agent output (right)
3
3
11
2,604