Jingang Wang

Ph.D.
Research Scientist at Meituan
Recursive Self-improvement & AI for AI
Opinions on my own

[小径残雪 @xiaojingcanxue]

耿同学质疑所谓的俄罗斯工程院院士是水货，然后所谓的院士带着3个人找上了门。

这一段是现场实录，很精彩。

[Jingang Wang @bitwjg]

@natolambert MOPD

[Jingang Wang @bitwjg]

@natolambert @allen_ai congrats nathan❤️

[Jingang Wang @bitwjg]

6/ RSI is getting more practical. Near-term self-improvement probably looks like: auto-generate tasks, sample failures, improve harnesses, select RL data, update skills, and run self-verify -> self-train loops.

[Jingang Wang @bitwjg]

May takeaway on Coding Agents and AI-for-AI:
the frontier is moving from "which base model is best?" to "who has the best training data, verifier, harness, and eval loop?"

[Jingang Wang @bitwjg]

5/ Harness became a core asset. LiteCoder-Terminal, unix-ctf, and agent failure taxonomies all point to the same bottleneck: better coding agents need better executable environments, not just better prompts.

[Jingang Wang @bitwjg]

4/ The data unit is changing. For agent SFT/RL, a "successful patch" is too coarse. The useful unit is the state transition: what the agent saw, tried, verified, failed, recovered from, and changed.

[Jingang Wang @bitwjg]

3/ RL post-training matured fast. May moved from rollout efficiency and cold-start tricks, to GRPO/credit assignment, to RL data selection: SAERL, label-free RL, verifier-free rewards, and offline RL for code.

[Jingang Wang @bitwjg]

2/ Eval is fragmenting. SpecBench, AgentLens-style trace diagnostics, FastKernels, TerminalBench-related work, DeepSWE, and SWE-bench Multimodal all point to one thing: final pass/fail is not enough.

[Jingang Wang @bitwjg]

1/ SWE-bench looked stable at the top in May. Public Verified scores hovered around 79%, while new work pushed toward harder evals: multimodal SWE, terminal tasks, production kernels, long-horizon agents, and safety.

[Jingang Wang @bitwjg]

@DhruvJain08 @thsottiaux YES， PLEASE！！！

[Jingang Wang @bitwjg]

@op7418 看来经常reconnecting是个普遍问题，我一直以为是我自己梯子的问题。还有就是自动压缩上下文的时候总是remote连接超时的问题

[Jingang Wang @bitwjg]

A worthwhile read. "The Ultra-Long Context Paradox" by Chen Zhang, the core contributor of Longcat Zigzag Attention. https://t.co/saps520U1A

[Chen Zhang @ChenZhang0212]

"Long context is a solved problem." It isn't.

Context rot. Context anxiety. Lossy compression that compounds over hours.

I wrote about why the answer is to contine scaling — it's to push past 1M tokens.

https://t.co/8BR599To0C

[Chen Zhang @ChenZhang0212]

"Long context is a solved problem." It isn't.

Context rot. Context anxiety. Lossy compression that compounds over hours.

I wrote about why the answer is to contine scaling — it's to push past 1M tokens.

https://t.co/8BR599To0C

[Jingang Wang @bitwjg]

@lorenlugosch Jason was probably constrained by a non-compete agreement, so he couldn’t use his real name.

[Jingang Wang @bitwjg]

@tvytlx 目前没有共识，各个harness框架都有不同的抽象和实现

[Wenhao Chai @ CVPR 2026 @wenhaocha1]

Introducing MLS-Bench for machine learning science.

Auto research built on coding agents is undoubtedly another major market beyond SWE coding. It is harder and more challenging. However, we believe there are two different categories here. Auto research from @karpathy , MLE-Bench, and PostTrainBench are one type of attempt: engineering. Agents are asked to optimize a specific engineering objective, but we do not require them to produce transferable, generalizable behavior.

MLS-Bench contains 140 tasks across 12 domains. Each task requires an agent to improve a specific component of an ML system or algorithm, and to demonstrate that the improvement generalizes and scales under controlled settings. We find that current agents are still far from consistently outperforming human-designed methods, and that engineering-style tuning is much easier for them than genuine method invention.

@Lyubh22 is the lead of this project. I was deeply impressed by the way he used Discord to organize agent trajectories and share them with the team.

Paper: https://t.co/pr2DgBcQto
Code: https://t.co/mvoirci6o6
Website: https://t.co/jmJMwfgChL

[Jingang Wang @bitwjg]

@natolambert Thanks for your visiting

[Jingang Wang @bitwjg]

@natolambert I think it is also feasible to update the knowledge cutoff time during the CPT/mid-training phase.

[Denny Zhou @denny_zhou]

Intelligence will not become a commodity.

Intelligence will not be too cheap to meter.

In the future, the real gap between people will be defined by their access to intelligence.

Just as a car lets an average person outrun any world champion, AI lets ordinary minds solve problems the greatest minds of the past could not.

Be prepared for this revolution.