Ting Sun

engineer at Wizard Quant

[Ting Sun @Sunt_ing]

great work!

[Deyao Zhu @tikgiau]

Introducing EdgeBench, a benchmark designed to study how agents learn from environments over at least 12~72-hour runs. We find that performance follows a log-sigmoid function of environment interaction time with high precision.

EdgeBench is built with three ingredients:

- 🌍 Real & Diverse: 134 real-world tasks across 6 task categories, spanning scientific problems, professional knowledge work, software engineering, optimization, formal math, and games.
- ⏳ Ultra-Long-Horizon: Each task supports 12–72 hours of agent work. Recorded human effort averages 57.2 hours.
- 🔁 Informative Feedback: Agents receive real-world feedback for continuous improvement.

After 38,000 hours of agent runs on EdgeBench, a scaling law for learning from environments emerges:

- 📈 As agents interact with task environments over time, their aggregate performance is precisely fit by a log-sigmoid function.
- 🧠 This phenomenon can be explained by an elegant theory of graph exploration.

We are releasing an initial 51 of the 134 tasks, together with the full evaluation framework, to help advance long-horizon agent research. Check our blog & paper for more findings!

Blog https://t.co/nMOzFsOhbT
Paper https://t.co/rZb3eWuvik
GitHub https://t.co/oemXd4UrFw
Dataset https://t.co/P4SQMrM47o

Details below 👇🧵

[JunjieZhang @JunjieZhang123]

🚀 We just cracked the code on autonomous AI agents!
Introducing L0: A breakthrough system that teaches LLMs to become true general agents through Reinforcement Learning with Verifiable Rewards (RLVR) 🧠
GitHub: https://t.co/cmgNv50CIR
Paper:
https://t.co/poJTBjI1LT

[Ting Sun @Sunt_ing]

@Xinyu2ML @iclr_conf it reminds me of cacheblend

[Ting Sun @Sunt_ing]

@Nils_Reimers @yoavgo do you know InferCept paper from ucsd? it tries to solve what u want to solve

[Ting Sun @Sunt_ing]

@tianyin_xu @IllinoisCDS @UMichCSE @MichiganSystems @ACMSIGOPS is this award based on looks?

[Cunxiang Wang @CunxiangWang]

🚀 Check out our new paper, "SHIELD: Evaluation and Defense Strategies for Copyright Compliance in LLM Text Generation"!
@_XiaozeLiu @Sunt_ing

🌐 We have developed a comprehensive framework to address copyright challenges in LLMs, (1/3)

[Ting Sun @Sunt_ing]

@agrawalamey12 it reminds me of Andes

[Ting Sun @Sunt_ing]

@MarisaVeryMoe Not a blog but a related keynote: "It's Time for Operating Systems to Rediscover Hardware" by Timothy Roscoe.
link: https://t.co/17sHIkJs4U

[Ting Sun @Sunt_ing]

@Guanhua_Wang_ Big cong!

[Ting Sun @Sunt_ing]

@Hackin9 Congratulations!

[Ting Sun @Sunt_ing]

@yyw2000 @Hackin9

[Ting Sun @Sunt_ing]

@MarkCallaghanDB How about "Scalable SQL"?

[César A. Hidalgo @cesifoti]

Want to read an insane tenure denial story? I was denied tenure at MIT in 2018. I was the only Hispanic faculty at my department, had 13k+ citations, 2 books, & papers in Nature, Science, & PNAS. I was never given a reason & the only letter I got told me to check the website./1🧵

[Ting Sun @Sunt_ing]

@vkalavri @easyAbi @pentium_ii Congratulations! 👏👏👏

[Ting Sun @Sunt_ing]

@Zhiqiang_Xie @awscloud @GraphDeep Congratulations!

[Ting Sun @Sunt_ing]

@pentium_ii Realized that it "was" hard for "me" only. So bro could you give me a recommended professor list privately? 🥰

[Ting Sun @Sunt_ing]

@vkalavri Hi, Ms Vasia. Thanks for your efforts in making the course. I'm wondering whether the materials (such as the course videos and slides) will be available for students outside of BU?

[Ting Sun @Sunt_ing]

@dpovey1 Wow, I'm looking forward to it! 🤪🤪

[Ting Sun @Sunt_ing]

@dpovey1 Hello, Professor Daniel Povey.
I come from SUSTech, China. I'm a junior student, majoring in CS. My classmates and I were very excited when we heard that you were going to work in China. We hope you can consider joining SUSTech.

[Ting Sun @Sunt_ing]

@dpovey1 true.