Fenghai Li

CS & Math @UIUC | AI for Scientific Discovery | Research Agents & Self-Evolving AI

[Fenghai Li @FenghaiLi7]

Research agents can already propose ideas and run experiments. Can they also tell us what researchers will work on next?

We built IdeaForecastBench to test this: give an LLM the literature available at a past cutoff, ask it to forecast research ideas, and compare them with papers published over the next three months.

Our EMNLP 2026 paper, “Can Large Language Models Forecast What Researchers Study Next?”, is out, along with the code and data.

📄 https://t.co/dEIBsX0Iq9
💻 https://t.co/gMgDJp4wy3
🤗 https://t.co/lELXBghrxj
A few things we learned 🧵

[Fenghai Li @FenghaiLi7]

@Dakshagarwal76 Yes!

[Fenghai Li @FenghaiLi7]

Great to see growing interest in this direction!

We explored a closely related question in our earlier #EMNLP2026 paper, CAN LARGE LANGUAGE MODELS FORECAST WHAT RESEARCHERS STUDY NEXT?

Our benchmark, IdeaForecastBench, evaluates LLM-generated research ideas against actual future publications.

Paper: https://t.co/9DiP5N7Xm1

[Shanshan Zhong @zhongshsh33]

AI agents can now run most of the research loop. But can they propose research worth doing?

We built IdeaTaste to find out:
📄 https://t.co/T3wjPz7bU3
🏆 https://t.co/oDXZqDbST4

[Fenghai Li @FenghaiLi7]

@Dakshagarwal76 Doing pretty well, we will release another paper really soon!

[Fenghai Li @FenghaiLi7]

@pengrui_han @CFCamerer super inspiring!!

[Pengrui Han (Barry) @pengrui_han]

Had a wonderful time giving a guest lecture at Caltech on the cognitive architecture of LLMs, and more broadly, how we can use AI models as model organisms to study intelligence! :)

Many thanks to Prof. @CFCamerer for the invitation!

We discussed the emergent modularity in LLMs, what the similarities between human brains and LLMs might tell us about general principles of intelligence, and how AI models could help us study cognition.

Really enjoyed the discussions and questions! :)

Slides: https://t.co/6s6sut0CD5

(And special thanks to @devarda_a for letting me steal some of his beautiful slides! :))

[Fenghai Li @FenghaiLi7]

@StevenyzZhang @Diyi_Yang @zsoltkira @kartik_goyal_ @PoloChau @sanmikoyejo Congrats Dr. Zhang!!

[Yanzhe Zhang @StevenyzZhang]

PhDone!

Very grateful to all my committee members!

@Diyi_Yang @zsoltkira @kartik_goyal_ @PoloChau @sanmikoyejo

[Fenghai Li @FenghaiLi7]

@Haosen9527 但是搞几个饮水机应该不难

[Fenghai Li @FenghaiLi7]

水文比水多

[Xuan (Billy) Zhang @xzhang_billy]

#EMNLP2026 vs. #COLM2026
1. EMNLP 的晚宴屈指可数，COLM 每天可以跑不止一场晚宴。
2. COLM 的 industry booth 数量远大于 EMNLP。
3. 在 EMNLP 会被渴死，但 COLM 不会。
4. EMNLP 离 Downtown 太远，而 COLM 就在 Downtown。
5. 湾区的中餐数量远大于Budapest。

#COLM >> #EMNLP Q.E.D

[Fenghai Li @FenghaiLi7]

@dontrustmebro 我是疯狂忍到coffee session💀

[Fenghai Li @FenghaiLi7]

@HaritzPuerto Emnlp should provide a good tts api 🤣

[Fenghai Li @FenghaiLi7]

@gneubig Sell the tokens and fund students Claude 20X

[Fenghai Li @FenghaiLi7]

Looking forward to meeting new people and chatting about exciting research ideas!

[Fenghai Li @FenghaiLi7]

AI can already conduct research. But can it predict what scientists will study next?

I’ll be presenting our work “Can Large Language Models Forecast What Researchers Study Next?” at #EMNLP2026 tomorrow!

📍 Room G13-14
🕓 Oct 25, 16:00–17:30 | Oral Session 4

Come say hi! Happy to chat about AI for scientific discovery and self-evolving agents!

https://t.co/dEIBsX0Iq9

[Fenghai Li @FenghaiLi7]

Not sure if this includes mine, but I'll be presenting our work on predicting what researchers will study next at #EMNLP2026 tomorrow! Come check out our oral presentation in Room G13-14, Oct 25, 16:00–17:30. Happy to chat about AI for scientific discovery and self-evolving agents! https://t.co/dEIBsX0Iq9

[Haofei Yu @haofeiyu44]

🤔️How to build sample-efficient RL algorithms for long-horizon language agents?

🚀Our answer: Just ONE additional loss on top of GRPO can achieve up to 2× sample efficiency across 5 long-horizon agent benchmarks!

By adding a multi-step action forecasting loss, we encourage the agent to learn action chunks from successful rollouts. Surprisingly, this simple idea works really well!

GitHub: https://t.co/TvvJY79wfp
Project page: https://t.co/Mh1hqT67sQ
Paper: https://t.co/mfS9nnCAfA

Here's the thread 🧵👇

[Sakana AI @SakanaAILabs]

The Agentic Scientist: Orchestrating Coding Agents for Automated Scientific (Co-)Discovery

Blog: https://t.co/7aVDvZd6m2
GitHub: https://t.co/JCCBMvu2P1
arXiv: https://t.co/4bV76tAByj

Since 2024, we’ve used The AI Scientist as a way to test how far AI alone can take scientific research, from v1 to v2 to the paper published in Nature earlier this year. The Agentic Scientist is the next step in that line of work, built for the era of coding agents.

Coding agents such as Codex and Claude Code now carry out long software tasks on their own. Handing them a research problem, however, tends to produce a strong hyperparameter optimizer rather than a scientist: scores go up, but ideas stay narrow, and the reasons for the gain are rarely examined.

The Agentic Scientist makes the research loop explicit. It wraps off-the-shelf coding agents in a scaffold that controls what they see, what they write, how they are evaluated, and how results are handed to the next stage.

Parallel workers test hypotheses against rubrics defined before any experiment runs, share successes and failures through a common memory, and then ablate the winning program to generate evidence.

We used an early version of the system to study Error Diffusion, a biologically motivated alternative to backpropagation. Our researchers set the question and the rubric, chose directions at each stage, and verified the results, references, and claims before submission. The resulting paper was accepted as an oral presentation at ALIFE 2026. In autonomous runs, the system also raised Doc-to-LoRA retrieval accuracy from 52% to 98% on multi-needle tasks and designed a learned evolution strategy that improves over its CMA-ES reference on unseen tasks.

Alongside it we release AutoIllustrate, a pipeline for scientific figures that keeps an editable https://t.co/SHX9lrOr74 file as the source of truth, so every diagram can be corrected by hand.

Humans define the question and the standards of evidence; agents expand the experiments researchers can run. We call this Co-Discovery, and we are releasing both systems as open source for the community to build on.

[Yanzhe Zhang @StevenyzZhang]

You have to have a scientific harness to make scientific progress.

This is the one. 🔥

[Fenghai Li @FenghaiLi7]

@SophieLWang @RulinShao Very useful especially in this ai era. Thanks a lot!