Vishnu Teja Kunde

Empowering and taming artificial brains using agentic harnesses | PhD @TAMU | Agentic AI + RL + Gen AI

[Vishnu Teja Kunde @sampleparticle]

@qizhengz_alex Great work pushing the frontier of automated research, especially in long-horizon settings. It'll be interesting to compare to our recent work AgentDiscover (https://t.co/0zxJUt81VD) which also accounts tokens and the "resetting" of the agents. Congrats 🎉

[Vishnu Teja Kunde @sampleparticle]

@diffusion_llms @COLM_conf @ssahoo_ @zhihanyang_ Looking forward to it! Please check out our work on RL for diffusion language models (https://t.co/RJKQEjL2ji) which is being presented in Non-AR LM workshop at COLM2026 on Oct 9 (8:30 AM). This work is appearing in NeurIPS 2026 as well.

[Vishnu Teja Kunde @sampleparticle]

@RulinShao @shannonzshen @oscar_yinn @yuetai12575 @hamishivi @natolambert @TengX6 @ml_perception @scottyih @LukeZettlemoyer @PangWeiKoh Great work, @RulinShao! Our recent research was in agent driven context management and this is very inspiring. Thank you!

[Vishnu Teja Kunde @sampleparticle]

Thrilled to see our work out! We present a principled and efficient RL algorithm for Diffusion LLMs, with strong results on diverse reasoning benchmarks.

Thanks to all collaborators!

[Fatemeh Doudi @Fatemehdoudi]

1/n New paper🗞️:

🚀 How do you do RL post-training for diffusion LLMs the right way?

Introducing EGSPO & EGSPO-SA — a principled RL framework for diffusion language models that achieves SOTA on coding & logical reasoning benchmarks 📷

w/ @sampleparticle,Mahdi Farahbakhsh, @DileepKalathil @krishna75095599, and Jean-Francois Chamberland.

📜 Paper: https://t.co/Z0HqvyKkrN
👩‍💻 Code: https://t.co/9fhdTIgftK
🤗 Models: https://t.co/TWCzT9otS8