X
Rishabh Agarwal

Rishabh Agarwal

Periodic Labs

1,664 posts

Rishabh Agarwal

Periodic Labs

@agarwl_

Reinforcement Learner

agarwl.github.io

Joined May 2016

RepliesReplies MediaMedia

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone

or

Log in with username or email

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.

Pinned
Rishabh Agarwal
@agarwl_
Apr 28
I gave a talk at ICLR 2026 about how we are scaling RL on frontier LLMs with 1T+ parameters, on experimental data from our physical lab at Periodic! Here's a rough recording of the talk:
00:00
Rishabh Agarwal
@agarwl_
Aug 24
Mujoco but real
Linus ✦ Ekenstam
@LinusEkenstam
Aug 24
I did not expect it to be THIS fun watching robot olympics 😂
00:00
Rishabh Agarwal
@agarwl_
Aug 18
Every follow-up OPD paper should be required to come with a video analogy
Kasey Zhang
@_WEEXIAO
Aug 17
how the student model probably feels during OPD
00:00
Rishabh Agarwal
@agarwl_
Aug 13
Annoying that sucking supervision bits with a straw works so well for LLMs
Rishabh Agarwal
@agarwl_
Aug 13
Good blog, makes you think about the empirical observation that cureent RL methods that work for LLMs are *low bias* - value functions trade off variance with bias, and hasn't shown huge gains yet - small bias from trainer-inference mismatch often is catastrophic for scaled up
beren.io
How can LLM RL Work Despite Information-Theoretic Inefficiency
Epistemic Status: Obviously speculative and maybe obvious. The success of RL in LLMs has been puzzling me for a while. People have developed various information-theoretic style arguments by which...

Advertisement

Advertisement