Zihan "Zenus" Wang ✈️ ICML(@wzenus):In Agent RL, models suffer from Template Collapse. They generate vast, diverse outputs (High Entropy) that lose all meaningful connection to the input prompt (Low Mutual Information). In other words, agent learn different ways to say nothing. 🚀 Introducing RAGEN-v2 -- Here's how we define and fix such silent failure modes in Agent RL. 🧵

Zihan "Zenus" Wang ✈️ ICML

@wzenus

Reasoning agent / RL / efficiency research @NorthwesternU & @nvidia. Ex @Microsoft @yutori_ai @deepseek_ai @uiuc_nlp @RUC1937.

Joined March 2022

687 Following 23K Followers

Zihan "Zenus" Wang ✈️ ICML@wzenus

2026.03.12 22:31

In Agent RL, models suffer from Template Collapse. They generate vast, diverse outputs (High Entropy) that lose all meaningful connection to the input prompt (Low Mutual Information). In other words, agent learn different ways to say nothing. 🚀 Introducing RAGEN-v2 -- Here's how we define and fix such silent failure modes in Agent RL. 🧵