Log inSign up
user avatar
Qihan Ren
@jsonren00
Ph.D. @sjtu1896. Prev. Undergrad @sjtu1896 and @Umich. Intern at @Minimax_AI and @Alibaba_Qwen post-training. XAI, LLM safety& reasoning, Agentic RL.
Shanghai, China
nebularaid2000.github.io
Joined July 2024
133
Following
56
Followers
RepliesMedia
  • Pinned
    user avatar
    Qihan Ren
    @jsonren00
    Apr 14
    [1/8] A prevailing narrative in post-training: SFT memorizes, RL generalizes. We revisit this for reasoning SFT and find that cross-domain generalization is NOT absent, but highly conditional, jointly depending on optimization, data, and base model capability. 📄 Arxiv:
    Image
    Image
  • user avatar
    Qihan Ren
    @jsonren00
    Feb 12
    Fast and brilliant
    user avatar
    MiniMax (official)
    @MiniMax_AI
    Feb 12
    Introducing M2.5, an open-source frontier model designed for real-world productivity. - SOTA performance at coding (SWE-Bench Verified 80.2%), search (BrowseComp 76.3%), agentic tool-calling (BFCL 76.8%) & office work. - Optimized for efficient execution, 37% faster at complex
    Image
  • user avatar
    Qihan Ren
    @jsonren00
    Feb 5
    Protect and understand your agent with our AgentDoG🐶! There's still much to be done in the future though...
    user avatar
    Dongrui Liu
    @dong_rui39501
    Feb 4
    [1/8] 🐶 Introducing "AgentDoG": A Diagnostic Guardrail Framework for AI Agent Safety. It achieves SOTA performance, diagnosing root causes (e.g., prompt injection, tool misuse) with 82% accuracy, far surpassing general LLMs. 📄 Paper: arxiv.org/abs/2601.18491
    Image