Yifan Kang @ CoRL 2026

MIT '28 | robotics @MIT_LISLab

[Xianbang Wang @kevinxbwang2007]

A bigger teacher should give you a better DMD student, right?
👉 NOT if your critic is small. The student gets worse, and can even collapse.

2 dead-simple critic fixes everyone overlooks:
1. Scale the critic with the teacher (joint scaling)
2. Tell the critic which generation step it's judging (STAC)

🌐 Project: https://t.co/GQT7XYSCWB
📄 Paper: https://t.co/WPPyWUCGPx

[Yifan Kang @ CoRL 2026 @kyiiiifan]

@MIT_LISLab I’ll present KeyFlow at CoRL’s Learn@Deploy workshop on Nov 12!

With Ben Zandonati (equal contribution), Tomás Lozano-Pérez, and Leslie Pack Kaelbling at MIT.

Come chat about robot-use agents and deployment-time adaptation!

https://t.co/Gh96ed0xVv (7/7)

[Yifan Kang @ CoRL 2026 @kyiiiifan]

Excited to share my first project at @MIT_LISLab : 𝐊𝐞𝐲𝐅𝐥𝐨𝐰! 🤖

Can robots adapt from a few demos without retraining? We let a coding agent improve a retrieval-based policy through simulated play.

Presenting at CoRL Learn@Deploy workshop! 🧵(1/7)
https://t.co/Gh96ed15L3

[Yifan Kang @ CoRL 2026 @kyiiiifan]

@MIT_LISLab Can KeyFlow also improve a robot foundation model?

In IsaacSim, its flow fields guide a 𝐟𝐫𝐨𝐳𝐞𝐧 π₀.₅ policy, improving mean success by 𝟑𝟕.𝟐 𝐩𝐞𝐫𝐜𝐞𝐧𝐭𝐚𝐠𝐞 𝐩𝐨𝐢𝐧𝐭𝐬 across six tasks—without retraining the base model. (6/7)

[Yifan Kang @ CoRL 2026 @kyiiiifan]

@YijiaFanCSAI What is the difference between this and Guava? https://t.co/iMdf3nwcA9

[Yifan Kang @ CoRL 2026 @kyiiiifan]

Very surprising results!😨

[Rikiya Takehi @rikiyatakehi]

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.

We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: https://t.co/bOWCcFhCVq

[Yifan Kang @ CoRL 2026 @kyiiiifan]

This is massive

[Ryan Williams @rrwilliams.bsky.social @rrwilliams]

An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4

[Yifan Kang @ CoRL 2026 @kyiiiifan]

Awesome class! I am taking it right now. Would recommend to people interested in learning about self-evolving LLMs

[Paul Liang @pliang279]

The next frontier is AI that improves itself, and we believe everyone should know how to build it.

This semester, Ao @ao_qu18465 and I are teaching 'How to AI an AI for Almost Anything' @MIT @medialab -- a new course on self-evolving AI systems that recursively improve from experience, feedback, memory, and exploration over long interaction horizons.

The key idea: intelligence shouldn't stop improving when training ends.

We cover four levels of evolution:
(1) context evolution through prompt optimization and context management,

(2) harness evolution through memory, tools & skills, workflow optimization, and multi-agent systems,

(3) model evolution through RL, distillation, test-time training, self-play, and continual learning,

(4) task evolution through generating increasingly complex synthetic tasks, simulations, evaluations, and environments.

Slides, readings, and videos available: https://t.co/QLPhT77Yvj

We thank @Kimi_Moonshot and @PrimeIntellect for generously sponsoring API credits!

[Yifan Kang @ CoRL 2026 @kyiiiifan]

@mt_saka How do you like it? I have not tried it before. I have had santouka which is not bad but not 二郎系

[Phillip Isola @phillip_isola]

Students sometimes ask me if it still makes sense, in this accelerating age, to pursue a PhD in AI.

Perhaps counterintuitively, I think it's a great time to do so.

I wrote up some thoughts on this here: https://t.co/RRqGR0qi1R

[Yifan Kang @ CoRL 2026 @kyiiiifan]

@fjzzq2002 I see. Still very cool!

[Benedict Quartey @Benedict_Q]

Leslie Kaelbling on why Astra is so good at robotics and why robotics still isn’t solved “yet”.

I spoke at Leslie’s lab pre-Astra, and we chatted about many of these ideas. Pretty cool to see them coming into the mainstream. #IROS2026

[Jiafei Duan @DJiafei]

Seeing at least 10-20 “agentic robotics” papers built around GPT-6 Astra + a robotics harness out this week.
But the core idea isn’t new. VoxPoser, Code as Policies, Manipulate-Anything, and others explored LLM-driven robot orchestration years ago, with even deeper roots in TAMP.

What has changed is the underlying technology.

[Yifan Kang @ CoRL 2026 @kyiiiifan]

@luoluo_ai lol is that Zian next to you

[Yifan Kang @ CoRL 2026 @kyiiiifan]

Hanhong is extremely talented! This kind of talent is hard to find in the job market nowadays. Hire him before anyone else does :)

[Hanhong Zhao @Hope7Happiness]

I'm looking for full-time Research Engineer / Machine Learning Engineer roles starting Jan or May 2027! I'm an MIT undergraduate (majoring in Physics + AI) researching generative modeling in Prof. Kaiming He's group. I'm especially interested in LLM pretraining/post-training and multimodal generation.