Hanze Dong's Blogs Research notes and essays by Hanze Dong on reinforcement learning, agents, and AI systems. https://hendrydong.github.io/blogs/ 2026-03-13T00:00:00+00:00 Hanze Dong https://hendrydong.github.io/ When Execution Is No Longer Human https://hendrydong.github.io/blogs/pages/agentgov.html 2026-03-13T00:00:00+00:00 2026-03-13T00:00:00+00:00 AI agents exercise execution without legitimacy. Beyond formal authority and operational power, agent-mediated organizations create a third kind of power: control over the architecture through which work gets done. Curate the Learning Signal for Reinforcement Learning https://hendrydong.github.io/blogs/pages/rl-ada.html 2026-02-25T00:00:00+00:00 2026-02-25T00:00:00+00:00 One gradient-first story for LLM RL: variance-minimizing sample allocation (GVM-RAFT), non-linear pass-rate objectives with adaptive sampling (Reinforce-Ada), and self-hinting to open the GRPO gate (SAGE).