Rujun Han

Senior Research Scientist @Google working LLM agents, post-training and distillation. NLP+ML Ph.D. from @USC_ISI. Ex-@AWS AI, ex-@federalreserve.

[Wenda Xu @WendaXu2]

If you are into OPD (on-policy distillation), please don't forget to check our Speculative OPD (Speculative Knowledge Distillation).

Idea: For each token, sample it from the student but discard if it teacher is unlikely to generate it (top-K), and resample from teacher. This dynamically switches between on-policy and supervised distillation.

[Gaotang Li @GaotangLi]

🧩 How can we assign fine-grained credit over long tool-use trajectories and let agents learn from past attempts in agentic reinforcement learning when rewards are no longer verifiable?

Excited to share RubricEM, an RL framework for long-form deep research agents that plan, search, use tools, and write reports without exact answer checks.

📖 Paper: https://t.co/t1tksq5g30

(1/n)

[Rujun Han @HanRujun]

Thrilled to see our work TTD-DR (https://t.co/4bWp9XO4QE) empowers Gemini Deep Research Max to become the SOTA DR product!

[Sundar Pichai @sundarpichai]

We are launching two powerful updates to Deep Research in the Gemini API, now with better quality, MCP support, and native chart/infographics generation.

Use Deep Research when you want speed and efficiency, and use Max when you want the highest quality context gathering & synthesis using extended test-time compute — achieving 93.3% on DeepSearchQA and 54.6% on HLE.

[Sundar Pichai @sundarpichai]

We are launching two powerful updates to Deep Research in the Gemini API, now with better quality, MCP support, and native chart/infographics generation.

Use Deep Research when you want speed and efficiency, and use Max when you want the highest quality context gathering & synthesis using extended test-time compute — achieving 93.3% on DeepSearchQA and 54.6% on HLE.

[Rujun Han @HanRujun]

Attending #EACL2026 and will present our paper https://t.co/IgqBeVNsGl at Findings Poster Session 6 at 11am on March 27. Please come by and ask any questions if you are around, and happy to discuss internship and full-time opportunities at Google Cloud AI Research.

[Rujun Han @HanRujun]

Paper: https://t.co/IgqBeVO0vT
Code/Data: https://t.co/GRS7Sfysys (details to be added soon!)

Huge thanks to @brunchavecmoi and other co-authors (@anmourchen, @ZifengWang315, @IHung_Hsu, @jun_yannn, @eunsolc, @tomaspfister, @chl260) for the amazing work!

[Rujun Han @HanRujun]

🚀 New paper on synthetic data generation for deep search agents (accepted to #EACL2026 Findings)!

We introduce SAGE, an agentic pipeline for generating high-quality, difficulty-controlled training data for deep search agents on a given corpus, using execution feedback.

[Rujun Han @HanRujun]

Transfer: Even though SAGE generates data using a fixed corpus (Wikipedia), we show agents trained on it can adapt to Google Searchat inference time without further training, though further research is needed in this direction.

[Yihe Deng @Yihe__Deng]

Supervised RL accepted to ICLR 2026! Thanks to all my co-authors for the great work and especially @IHung_Hsu :)

[Tengxiao Liu @TengxiaoLiu]

🏧Giving your agent unlimited tool calls doesn't make it smarter.
💡Why? It lacks 'Budget Awareness'!
Introducing Budget Tracker, a simple plug-in that enables more effective scaling behaviors: higher performance, lower cost.
Paper: https://t.co/aKm2Tzt1wx

[I-Hung Hsu @IHung_Hsu]

🧠🚀 Excited to introduce Supervised Reinforcement Learning—a framework that leverages expert trajectories to teach small LMs how to reason through hard problems without losing their minds. 🤯

Better than SFT && RLVR.

Read more: https://t.co/taEL8Vk4X5

#llms #RL #reasoning

[Yossi Matias @ymatias]

New research introduces the Test-Time Diffusion Deep Researcher (TTD-DR) framework, advancing AI agents for complex research: https://t.co/R2PldcoZbO

[Cecile Tamura @ceciletamura]

Featuring: Dr. @HanRujun of @Google in a deep dive w/ @ceciletamura of @ploutosai
https://t.co/WSHqnW09gJ

[Yumo Xu @yumo_xu]

Excited to share our #ACL2025NLP paper, "𝐂𝐢𝐭𝐞𝐄𝐯𝐚𝐥: 𝐏𝐫𝐢𝐧𝐜𝐢𝐩𝐥𝐞-𝐃𝐫𝐢𝐯𝐞𝐧 𝐂𝐢𝐭𝐚𝐭𝐢𝐨𝐧 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐟𝐨𝐫 𝐒𝐨𝐮𝐫𝐜𝐞 𝐀𝐭𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧"! 📜 If you’re working on RAG, Deep Research and Trustworthy AI, this is for you. Why? Citation quality is critical for trust, but current metrics are falling short. Let’s fix that!

🧵 [1/10]

[Shahriar Golchin @ShahriarGolchin]

Can many-shot ICL be cached and still tailored per test sample?

We make it possible. 💡

Excited to share that our paper, "Towards Compute-Optimal Many-Shot In-Context Learning," has been accepted to @COLM_conf!

Paper: https://t.co/lBf4z5VCHN

#COLM2025 #LLMs #AI #ICL

[Rujun Han @HanRujun]

We performed in-depth analysis to show that diversity of search directions and timely incorporation of retrieved information help research agents scale more effectively. Read our paper for more detailed analysis: https://t.co/tDXt8vxlsV

[Rujun Han @HanRujun]

Very excited to share the project I've been working on over the past several months! We proposed Deep Researcher with Test-Time Diffusion, a novel method to leverage iterative draft+revision to tackle complex questions demanding exhaustive search and reasoning.

[Rujun Han @HanRujun]

Further enhanced by self-evolving algorithms, we achieved SOTA results over a variety of research benchmarks with broad coverage of industry domains such as finance, technology, biomedical and legal.