Marko Medvedev

Check out our new paper on internalization: the process of gradually "absorbing" chain of thought computations during training. Our results show that internalization can work for problems that are computationally hard to learn directly. We carefully study method and task specific factors that determine internalization success. To learn more, see arxiv.org/abs/2606.20937. With @nikostsilivis @nirmitj_ @_rkomma Nati Srebro. @NYUDataScience @TTIC_Connect

Excited to share our new paper on tractable learning via Positive Distribution Shift! Link: arxiv.org/abs/2602.08907 Joint work with @_rkomma, @Elisabetta68885, Theo, @galvardi, and Nati.

What if this was our final tweet

@elonmusk Upsetting them equally or making them equally happy. Negative reinforcement learning does propagate faster tho.

For Twitter to deserve public trust, it must be politically neutral, which effectively means upsetting the far right and the far left equally