Junhong Shen

@JunhongShen1

@MicrosoftAI | PhD @mldcmu | BS @UCLA | Previously Interned @GoogleDeepMind @AIatMeta @MSFTResearch @DeterminedAI

Mountain View, CA

sjunhongshen.github.io

Joined September 2020

701 Following

2,246 Followers

Replies Reposts Media

Pinned
Junhong Shen
@JunhongShen1
Jun 2
Belated career update: I graduated from CMU MLD and have been working at @MicrosoftAI Superintelligence Team for 3 months, focusing on RL and coding agents. Deeply grateful to my advisor @atalwalkar for his guidance and support throughout my PhD. Excited for what’s ahead!
26
Junhong Shen
@JunhongShen1
Aug 12
Be sure to try it! Feedback welcomed!
Mustafa Suleyman
@mustafasuleyman
Aug 12
Our first reasoning model, MAI-Thinking-1, is built from scratch. Now available in Microsoft Foundry. Kudos to the team! More below.
2
Junhong Shen
@JunhongShen1
Jun 2
Also excited to share that MAI-Thinking-1, the model I’ve been working on for the past few months, has been released! microsoft.ai/models/mai-thi… Check out our technical report (microsoft.ai/wp-content/upl…), where we share so many insights into its development and training!
MAI-Thinking-1 | Microsoft AI
From microsoft.ai
3
Junhong Shen
@JunhongShen1
Dec 2, 2025
I'm in San Diego for #NeurIPS2025 Dec 2-7! Will present two posters: 1️⃣ Thinking vs. Doing: Agents that Reason by Scaling Test-Time Interaction: We introduce an RL algorithm that scales the number of interaction steps to enable agents to learn richer behavior patterns.
arxiv.org
Thinking vs. Doing: Agents that Reason by Scaling Test-Time Interaction
The current paradigm of test-time scaling relies on generating long reasoning traces ("thinking" more) before producing a response. In agent problems that require interaction, this can be done by...
5
Junhong Shen
@JunhongShen1
Sep 19, 2025
Excited to share two papers accepted to #NeurIPS2025 ! 1️⃣ Thinking vs. Doing: Agents that Reason by Scaling Test-Time Interaction We introduce TTI, an RL algorithm that scales the number of interaction steps beyond thinking tokens per step. Our agents learn to act
10

Junhong Shen

Junhong Shen’s posts