Zhao Chenyang / 赵晨阳

Zhao Chenyang / 赵晨阳 appears in 177: 详解Kimi K3：强到冲击Anthropic估值的模型什么样？ as the infrastructure-side guest explaining [[KimiK3|Kimi K3]]. The source identifies him as a founder/member of [[RadixARC|Radix ARC / Ridix ARC]] and uses his comments to connect K3’s architecture to serving, cache behavior, kernel design, agent environments, and open-weight release boundaries.

His main contribution is the infra interpretation of why K3 is not only a benchmark event. He argues that [[KimiDeltaAttention|KDA]] changes prefix reuse and speculative-sampling rollback, that Kernel Development Agents make kernel work an early verifiable self-improvement loop, and that AgentIn shows how stronger isolation can safely expand agent permissions.

Connections

• Kimi K3, Kimi Delta Attention / KDA, Prefix Caching, and AI Inference Cost Structure — serving and architecture topics he explains.
• Kernel Development Agents, CUDA, Moore Threads / 摩尔线程, and Domestic AI Chip Catch-Up — kernel and chip-adaptation branch.
• AgentIn, Agent Environment Isolation, Agent RL, and Model Harness Co-Evolution — agent-training environment branch.
• Open Weight Release Boundary, Open Source AI Models, and Open Model Safety Governance — open-weight and governance branch.