News

• 10/2026 🌋: Check out our new work Personal-Agent Mediated Recommendation, which trains personal agent to make better recommendations with cross-platform user history!

• 08/2026 🌋: Our work Agentic Chain-of-Thought Steering accepted to EMNLP 2026 Findings, which builds a controller agent that steers reasoning models in-flight for efficient and budget-aware thinking! See you in Budapest!

• 05/2026 🌋: Our work Simultaneous Multi-objective Alignment is accepted to ICML 2026!

• 04/2026 🌋: Check out our new work Hint Learning for Reinforcement Learning, which learns adaptive, transfer-aware hints for RL that improve reasoning in no-hint settings!