Xi Wang

'25 NYU CS PhD, AI for Science, Generative AI.
Stanford Visiting Scholar, ByteDance Seed.
Prev: SJTU, DP Technology, Microsoft.

[Xi Wang @EscheFlex]

@dwgreyman Replica exchange?

[Xi Wang @EscheFlex]

@AndyLin2001 Will your team focus on general auto research or put really application on some specific topic, like material/drug design?

[Xi Wang @EscheFlex]

@shi_weiyan What is more, we might think that people with training may already know how to make a “bomb” but the truth is they can’t do such things that easy. AI can greatly boost such process

[Xi Wang @EscheFlex]

@shi_weiyan Making bomb is more like a general idea, common sense level. The interdisciplinary AI safety needs some basic training on such disciplines, like undergraduate students. What people don’t aware is that current ai safety system is very vulnerable for basic to intermediate experts.

[Xi Wang @EscheFlex]

@BetaTomorrow @readwise save

[Xi Wang @EscheFlex]

@vikingmute @readwise save

[Xi Wang @EscheFlex]

@liu730chaoqi @LeRobotHF @readwise save

[Xi Wang @EscheFlex]

@vai_viswanathan @readwise save

[Xi Wang @EscheFlex]

@berryxia @readwise save

[Xi Wang @EscheFlex]

Introducing the Lab from the Future!

[AINORO @ainoro_labs]

What can a dexterous robot do in your wet lab?

Announcing AINORO’s partnership with LabOS @AI4S_Catalyst.

Together, we are exploring a workflow where AI can reason, build protocols, and execute experiments through AINORO’s dexterous robots.

From idea → protocol → robotic execution.

Demo below 👇
@ainoro_labs @AI4S_Catalyst

[AINORO @ainoro_labs]

20 years ago, lab automation meant: load the plates, press start, walk away.

In 2026, AINORO is bringing that simplicity to the entire wet lab.

Our affordable dexterous robot uses existing instruments, adapts to current lab environments, and helps automate workflows without rebuilding the lab.

The future lab will have robots working side by side with scientists — and with each other.

Introducing AINORO.

Meet us at https://t.co/FoQgTLcLZt

[Xi Wang @EscheFlex]

@Charles_Y_Wu @ainoro_labs Amazing work.... Robots will free scientist from tedious and repeated work😆

[Xi Wang @EscheFlex]

The arXiv version is still being updated:
https://t.co/M9eD10ghVR

[Xi Wang @EscheFlex]

Introducing RapTB ICML2026🎉

Most RL post-training methods are fundamentally mode-seeking: they optimize expected return. In sparse or binary-reward settings, methods like PPO/GRPO can easily collapse to a few high-reward modes, losing solution diversity.

Our work brings GFlowNets to LLM post-training.

Instead of training the model to find one maximum-reward trajectory, we train it to sample solutions with probability proportional to reward. This encourages broad exploration of the solution space, enabling the model to generate samples that are both diverse and high-reward.

Two key contributions:

1. Rooted-prefix training
Improves credit assignment and reduces trajectory-level variance.

2. Submodular replay selection
Lets the model automatically select replay samples that jointly balance diversity, reward, and length, reducing the risk of getting trapped in local modes.

We evaluate on tasks where diversity matters: molecule generation, AMP design, Game of 24 expression synthesis with binary rewards, and natural-language tasks.

Compared with PPO/GRPO and GFlowNet baselines such as TB/SubTB, we observe substantial improvements.

[Xi Wang @EscheFlex]

Robots are moving into wet-labs! Free scientists from laborious and tedious works.

[AI4Science Catalyst @AI4S_Catalyst]

The next generation of biology companies will run 24/7.

AI Scientists never sleep.

More soon. @ainoro_labs

[Xi Wang @EscheFlex]

@BoWang87 @naturemethods Congratulations! Nice paper. Have you tried to predict alternative splicing events? It’s way more complicated and difficult, but really meaningful to disease prevention

[Charly Wargnier ♨️ @DataChaz]

Explore here → https://t.co/wQ6sq1BPmy

Great job @Charles_Y_Wu @EscheFlex @wenbolu @JinglinJian @LingYang_PU 👏👏👏

Under the mentorship of
@lecong
@MengdiWang10
@auto_bio_ai
@AI4S_Catalyst
@Stanford
@Princeton

[Xi Wang @EscheFlex]

@edwardjhu @OpenAI @Yoshua_Bengio @Mila_Quebec Also working on Gflownet+LLM: https://t.co/M9eD10ghVR , really like your previous work in ICLR24

[Xi Wang @EscheFlex]

@indie_maker_fox 找一个好点的ai绘图吧