Fengqing Jiang

PhD Student @UW, working on LLM research | Now @Snowflake | Prev @MSFTResearch @IntelAI @AmazonScience @UofIllinois @ZJU_China| Open for collaboration.

[Fengqing Jiang @fengqing_jiang]

@thsottiaux please develop a skill so I can let my codex to run claude code for those 'simple' task in subagent without wasting my claude plan while I stay with codex : )

[Fengqing Jiang @fengqing_jiang]

New GPT model is incredibly fast and so good to work with (with so much reset usage these days). If you also stuck with the slow Claude throughput, try codex : )

[Fengqing Jiang @fengqing_jiang]

@thsottiaux The more codex usage is, the faster it glows. So please reset it (or bank it) : )

[Fengqing Jiang @fengqing_jiang]

At ACL!

I will present our paper poster at 11am on July 7th! The spot is L1 G130.

Welcome to catch me for chat!

#acl #acl2026

[Zhangchen Xu @zhangchen_xu]

🚀 We built Visual Aesthetic Benchmark (VAB). Arena is alive here: https://t.co/GKSvoJ5jAk

Aesthetic judgment is one of the hardest ceilings for AI to crack right now. Not generating images, but truly understanding what “looks good.”

We hand-curated 400 sets of artist works (fine art, photography, and illustration), featuring 2000+ hours of brand-new commissioned data created specifically for this benchmark — all grounded in 13K+ domain expert judgments across 7 core aesthetic dimensions (composition, lighting, technique, expression…) to ensure rigorous evaluation in highly subjective domains.

We asked 20+ frontier AI models to judge visual aesthetics (fine art, photography, illustration) against domain experts.
Frontier models are really not good at it yet.
Best model, Claude Sonnet 4.6 hit 26.5%. Human experts: 68.9%.

> Blog: https://t.co/XvEgwkmRqr
> Leaderboard: https://t.co/BEsHCgTmjC

[Fengqing Jiang @fengqing_jiang]

@micahgoldblum Sounds like just our BadScientist work indicates LLM-generated paper can easily fool LLM reviewers.

https://t.co/fwdc2MusHi

[Fengqing Jiang @fengqing_jiang]

Thanks for all collobrators, @YichenFeng_, @yuetai12575, Luyao Niu, @BaselAlomair, @RadhaPoovendran.
Sincerely appreciate the general support from the organizer #agents4science and @togethercompute.

[Fengqing Jiang @fengqing_jiang]

“Research on AI, by AI, for AI and all, shall not perish from the scientific community.”

Thrilled to share our new work BadScientist—honored with the Best Paper Award at Agents4Science 🏆.

[Fengqing Jiang @fengqing_jiang]

“Research on AI, by AI, for AI and all, shall not perish from the scientific community.”

Thrilled to share our new work BadScientist—honored with the Best Paper Award at Agents4Science 🏆.

[Fengqing Jiang @fengqing_jiang]

#AIResearch #AIAgents #LLM #ResponsibleAI #GenAI #OpenScience #Evaluation #AISafety #Agents4Science

[Fengqing Jiang @fengqing_jiang]

“Research on AI, by AI, for AI and all, shall not perish from the scientific community.”

Thrilled to share our new work BadScientist—honored with the Best Paper Award at Agents4Science 🏆.

[Fengqing Jiang @fengqing_jiang]

In parallel with our prior work SoSBench (https://t.co/3Mo28Mnjr0) within AI4Science, we aim to contribute to more rigorous AI-accelerated science. Feedback and collaborations welcome!

[Fengqing Jiang @fengqing_jiang]

🎉 Thrilled to share that I joined MSR RiSE Group and AI Red Team as research intern, working on LLM/Agent safety research. ✨

[Fengqing Jiang @fengqing_jiang]

🎉Thrilled to share our work got best honorable mention at ICLR BiAlign @bi_align workshop. Really appreaciate the great opportunity and thanks for the collaborative work of all authors.
#ICLR2025

[Fengqing Jiang @fengqing_jiang]

1/6 🤖
Introducing SAFECHAIN—a new study on Large Reasoning Models (LRMs) with lengthy chain-of-thought (CoT). Key takeaway: more reasoning ≠ automatically safer answers. #AI #NLP #LLMSafety

[Fengqing Jiang @fengqing_jiang]

While in our SafeChain paper (released on 2025/02/17), we explored similar idea first as well as LessThink and MoreThink. If you are interested in the effect of such design, don't miss it.

https://t.co/dctnLBRxd7

[Fengqing Jiang @fengqing_jiang]

Qwen 3 just dropped. A very interesting design in the system is the `enable_thinking` control in the chat template.
If it is disabled, then model will be enforced to an empty thinking trace. In other words, Qwen 3 is an intrinsic reasoning model
#Qwen3 #Qwen #LLM

[Fengqing Jiang @fengqing_jiang]

{%- if enable_thinking is defined and enable_thinking is false %}\n {{- '<think>\\n\\n</think>\\n\\n' }}

src from from tokenizer