BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CAIL//Events//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:CAIL Events
BEGIN:VEVENT
UID:2026-10-06-haonan-wang@cail.columbia.edu
DTSTAMP:20261013T005522Z
DTSTART:20261006T163000Z
DTEND:20261006T173000Z
SUMMARY:ML Seminar: Haonan Wang - Towards Recursive Self-Improvement: Less
 ons from Self-Evolving Reinforcement Learning
LOCATION:CSB 453
DESCRIPTION:Can language models sustain improvement by generating their ow
 n training tasks? This talk explores the promise of recursive self-improve
 ment through a systematic study of self-evolving reinforcement learning. W
 e discuss what current approaches achieve\, what our findings reveal about
  the limits of that progress\, and why the results so far do not establish
  a sustained cycle of self-improvement. We will also discuss possible dire
 ctions for addressing these limitations.\n\nhttps://cail.columbia.edu/even
 ts/2026-10-06-haonan-wang
URL:https://cail.columbia.edu/events/2026-10-06-haonan-wang
END:VEVENT
END:VCALENDAR