Luo Fuli / 罗福莉

Luo Fuli is the guest in 138. 对罗福莉3.5小时访谈：AI范式已然巨变！OpenClaw、Agent范式很吃后训练、卡的分配、组织平权. The source describes her as formerly at Alibaba DAMO Academy and DeepSeek, and as the current leader of Xiaomi’s large-model team working on the Memo VR series.

Her central claim is that the AI paradigm is shifting from chat products and pretraining-dominant competition toward Agent Post-Training, Agent Harness design, Agent RL, long context, tool use, and Model Harness Co-Evolution. She reads Open Claw and Open Cloud as agent frameworks that changed how her team thinks about research support, team management, user-agent data, skills, and workflow execution.

Source Position

• Luo treats agent frameworks as infrastructure between humans and models, not merely product interaction layers.
• She argues that top models and top agent frameworks will co-evolve, and that post-training must increasingly target agent workflows rather than only chat behavior.
• She presents Memo VR as a multi-model system where Pro, Omni, TTS, Flash, and long-context architecture choices are coordinated around agent scenarios.
• She describes a flatter AI Organization Design style at Xiaomi, with fewer rigid group boundaries and more movement across pretraining, post-training, infrastructure, and research problems.
• Her hiring frame emphasizes curiosity, love for the work, strong fundamentals, diversity, and environment over long prior experience.

Connections

• Xiaomi and Memo VR — current model-team and model-series context.
• Alibaba and DeepSeek — prior institutional background named in the source.
• Open Claw and Open Cloud — agent frameworks that changed her view of model training and team workflow.
• Agent Post-Training, Agent RL, Training Compute Allocation, and Agent-Optimized Model Architecture — model-training concepts introduced through her interview.
• Research Taste, AI Organization Design, and Model Harness Co-Evolution — human and organizational capabilities she treats as strategic in the agent era.