New paper: VLMs can self-reward during RL training — no visual annotations needed! -- Decompose VLM reasoning into visual vs. language parts -- Prompt the same VLM without visual input for visual reward We call it 𝐕𝐢𝐬𝐢𝐨𝐧-𝐒(𝐞𝐥𝐟)𝐑𝟏: arxiv.org/abs/2508.19652

❓ What if an LLM trying to self-improve without labels... actually makes itself worse? Check out our EVOL-RL!

Our MPS-Prover has been accepted by NeurIPS 2025. See you in SD~

Code for 𝐏𝐚𝐫𝐚𝐥𝐥𝐞𝐥-𝐑𝟏 is live! 👉 github.com/zhengkid/Paral… (now 189 stars and climbing 🔥) It lets LLMs think in parallel — multiple reasoning paths, smarter synthesis, more creative inference! Miss this paper and you’re missing a leap forward: arxiv.org/abs/2509.07980

More GPU means “better” foundation model (FM) research?? We looked at 6517 FM papers and surveyed 229 FM authors to understand the role of computing resources in publishing. Surprisingly….. arxiv.org/abs/2510.13621

I will be at ICCV 2025, happy to chat about our internship opportunities. See you at Hawaii🌴