Yangruibo Robin Ding

Want the agent to evolve themselves? You need an agent-centric ADK to empower them🤖🦾🦾!! Our agent-centric ADK, OpenSage, Available Now: 📄 Paper: arxiv.org/abs/2602.16891 🌐 Website: opensage-agent.ai 💻 Code: github.com/opensage-agent… 📝 Blog: rdi.berkeley.edu/blog/opensage/

Our new position just dropped. Given the rise of cybersecurity abilities of @AnthropicAI and @OpenAI, we argue that making AI safer may not be the way to make the world safer. We believe that the future is about training offensive AI agents to defend against cyber attacks.

Equipping SLMs as agents is hard, since established training recipes for “L”LMs do not work when model sizes are 10x smaller. But, we show it is not impossible — we need to carefully redesign the recipe specifically for “S”LMs. Check out our early explorations w/ SWE-Spot.

[1/n] 🚨 Coding ≠ Software Engineering! Are AI agents ready to replace Software Engineers? 🔥 Introducing DevOps-Gym: The first end-to-end benchmark for the complete software cycle (UCSB, NUS, Berkeley, Google). We tested SOTA agents on 700+ real-world DevOps tasks. The Result? They struggle. 📉 🔄 Full DevOps Coverage: 🔧 Build: Fix dependency hell & migrate systems (Maven→Gradle) 📊 Monitor: Detect leaks using ONLY CLI tools (top/iostat). 🐛 Fix: Resolve bugs in compiled langs (Harder than Python!) ✅ Test: Gen regression tests from runtime behavior ☠️ The Ultimate Killer: End-to-End Pipelines (Build → Monitor → Fix → Test) Success Rate: 0.00%. NO agent could complete the full loop. 🔗 Check out the full research & dataset: devops-gym.com 📄 Paper: arxiv.org/abs/2601.20882

@yiling__LOU @siebelschool Congratulations, Yiling!

Thrilled to announce that I'll be joining UIUC CS @siebelschool as an Assistant Professor in Spring 2026! 📢 I’m looking for Fall '26 PhD students who are interested in the intersection of Software Engineering and AI, especially in LLM4Code and Code Agents. Please drop me an email if you are interested in working with me.

🚨🚨New tool alert! Building on top of our PatchPilot agent framework, Co-PatcheR further trains 3*14B specialized reasoning models for issue resolving. We propose the concept of collaborative patching, where three small models work together in an agent framework. This idea was inspired by the typical workflow of how humans resolve code issues (i.e., collaboratively work together with specific focuses). We believe that if we want to push the limit of small models, having multiple specialized models may be easier than having one model for everything. #LLM #agent4swe #LLM4code

📢 Excited to share our new paper: Challenges and Paths Towards AI for SWE We discuss: 🛠️ 6 sub-tasks needed for SWE 🤖 9 challenges of today's AI in SWE 🔮 9 future directions to address the challenges w/ collaborators from MIT, Berkeley, Cornell, Stanford, and UPenn ⬇️ (1/n)

@devanbu @baishakhir Thanks, Prem, for introducing SemCoder! We found that teaching LMs to perform "monologue" is surprisingly effective at execution reasoning and potentially help transfering the knowledge to more tasks. We look forward to seeing more interesting discussions along this direction!

Really interesting work, branded "SemCODER" from @baishakhir @RobinDing3 et al... core idea: with additional training on programs with rubber 🦆 explanations of traces, a small-ish pre-trained LLM can outperform GPT3.5. Surprising! arxiv.org/pdf/2406.01006

Introducing SemCoder, a semantic-aware Code LLM excelling in code generation and execution reasoning. Trained with high-quality data and novel way of aligning execution, only 6.7B model is outperforming GPT3.5 and CodeLlama 34B. link: arxiv.org/pdf/2406.01006 #LLMs, #AI4Code

🚨 #GPT4 doesn't understand the code/specification written by itself!? 🚨 🥳 Check out our #ICLR2024 paper "Beyond Accuracy: Evaluating Self-Consistency of Code Large Language Models with ldentityChain" 🥳#LLM Paper: arxiv.org/abs/2310.14053 Code: github.com/marcusm117/Ide… [1/6]

Exciting times at #NeurIPS2023 ✨DM me for coffee and chat on ML for code and beyond! Come see our CrossCodeEval poster (#522) on Tuesday from 10:45-12:45. Stop by and say hi!

Retweet appreciated!

@saikatch107 @NeurIPSConf Thanks Saikat!

@RobinDing3 @NeurIPSConf Congratulations Robin!

🚨Is your Code LM good enough at understanding the cross-file context when completing the program? Come and challenge it!! We will present a diverse and multilingual benchmark, CrossCodeEval, at @NeurIPSConf 2023! See you in New Orleans!

Hi, Everyone, I will be on the academic job market for 2024 Fall, with a research focus on Systems for Deep Learning and GPU-based Parallel and Distributed Programming. Feel free to reach out if your department has open positions. Thanks!

There are still 5 days to submit to the Annual Symposium on Machine Programming (MAPS). MAPS is fully sponsored by NSF this year with travel support for US-based students. Looking forward to your exciting work! mapsworkshop.github.io

MAPS (mapsworkshop.github.io) deadline is approaching soon. There will be travel support by NSF for the US students who will be attending the conference. Submit your work if you work in AI4Code area. @FSEconf

Missed the ICSE ddl? If your work is about ML+SE/PL, please consider the 7th Symposium on Machine Programming (MAPS) mapsworkshop.github.io MAPS is non-archival this year, so you can still submit it to other venues, e.g., ESEC/FSE'24. And you will get reviews before FSE'24 ddl!

We are organizing MAPS Workshop this year, co-located with ESEC/FSE 2023 in San Francisco! Consider submitting your exciting work related to machine programming for initial feedbacks! The deadline is Aug. 15.

Happy to share arxiv.org/pdf/2306.03234… received distinguished paper award @issta_conf. @RobinDing3 will present the paper at 1:30pm PDT (Deep-Learning for Software Analysis session) ... if you are around please come and join us.