PhD student @CILVRatNYU. Intern @MSFTResearch || Pre: Student Researcher @GoogleDeepMind , @Meta, @Mila_Quebec, @mcgillu || RL, ML, Neuroscience.
Manhattan, NY
Joined August 2017
- Tweets138
- Following379
- Followers706
- Likes1.6K
Pinned Tweet
Why should a generative model be diverse? 👀
If there’s an obviously best answer, why put probability on worse ones?
New paper: a perspective on what diversity is *for*, and an algorithm on post-training for the right amount 🌹.
🧵🔽 [1/12]
Anthony GX-Chen retweeted
I agree that options/hierarchy is one of the bigger open problems. We made some progress on scaling HRL with options recently: arxiv.org/abs/2509.00338. There the abstraction was in the form of different reward functions (given as priors), there could be other ways though
Anthony GX-Chen retweeted
i was diagnosed with thyroid cancer many years ago purely by chance, because i had to get blood work for my US green card application (thank you, USCIS!). i was in my mid-30’s and was reasonably healthy (or so i believed.) this was shocking and totally unpredictable news to me. thanks to the amazing team of clinicians at NYU Langone, i successfully went through a total thyroidectomy followed by radioactive iodine therapy. but, it was truly the most uncertain and confusing period of my life.
this is when i first realized that no one can opt out of healthcare. being sick isn’t what you choose to or choose not to. it is something that unpredictably happens to all of us at some point for one reason or another. this is why it is a healthcare system that we collectively build and use to support each other at the society level. because it must support everyone, the complexity of creating, maintaining and improving this system is a Herculean effort that calls for all the help in the world.
today, i am announcing @OrtetAI together with my dear co-founders, @keunwoochoi, @HenriDwyer, @elmanmansimov, @causalclaudia and Jeff, to lend our support to improving the health of each and every one of us. Ortet’s goal is to create a technological foundation that will enable health systems to deliver predictably better care to each and every patient. this will only be possible by working closely with all stakeholders in the broader ecosystem, which is enabled in part by Ortet’s close partnership with Thoreau.
Ortet is taking the very first step of a long journey ahead, and at the end of this journey, i strongly believe we would be able to say that Ortet had contributed to building better, more predictable healthcare and thus to improving health of each and every one of us.
see ortet.ai/ for more!
Anthony GX-Chen retweeted
Hot damn!
Congrats on the AGI, @OpenAI! 😉
kenforthewin.github.io/blog/…
And see this concurrent work which shows the ROSA+Max / ReMax set-level objective also works nicely for image diffusion models!
More evidence that this is a general objective that works across many settings :)
Nice visualizations! And thanks for citing our ReMax that coincides in the max setting. We actually had another concurrent work that uses this objective for diversity in image generation RL post-training:
arxiv.org/abs/2607.14962
The figure below shows one result on promoting color diversity with a distribution over reward functions and optimizing for the max over multiple samples.
First author: @Ku_Onoda
Anthony GX-Chen retweeted
Check out our latest work on how to view diversity in generations from a new lens of "distribution of reward functions" in RL.
Lovely time hosting @AntChen_ at GDM Montreal last summer and wonderful collaboration with Mark Rowland and other colleagues.
Finally, this was an amazing collaboration with @GoogleDeepmind 's RL folks. Some of the first RL papers I read was from them & working with them have really elevated my thinking as a researcher :)
@ankit_s_anand @ghitzah @shibl @eseraygun @SmallingDa72711 & more
[11/12]
📄 Paper: theory, analysis, and in-depth connections to related areas
arxiv.org/abs/2606.03962
📝 Blog: intuitions, code, and some fun interactive animations if you want to play with ROSA and try it in your own project
im-ant.github.io/rosa/
[12/12]
Anthony GX-Chen retweeted
LLMs fall for confirmation bias: they test evidence that confirms their guess, not evidence that could break it.
NYU’s CS MS alum Ayush Rajesh Jhaveri (@arhjhaveri), with CDS’ @eunsolc, @sucholutsky & @AntChen_ show a fix built for humans helps.
nyudatascience.medium.com/ll…
Hi friends! I am in Seoul 🇰🇷 attending #ICML2026 . Generally interested in RL and the right ways to post-train to get the policies we actually want.
Looking forward to catching up with old and new friends! Do reach out ☕️
I’ll also be presenting at the DEMO workshop on why policy gradient is fundamentally a bad 🙅🏻♂️ approach for RL post-training 👎, and an alternative direction forward 🚀💎
🕛 Sat, Jul 11, 2026
📍 ASEM Ballroom 203
openreview.net/forum?id=EV9w…
