DeepSeek Harness: When the Agent Loop Itself Becomes a Plugin
中文摘要
DeepSeek R1推理模型突破：其代理循环（思维链）能作为可复用插件，而非仅是基准数据。
English Summary
DeepSeek's R1 reasoning model breakthrough: its agent loop (chain-of-thought) functions as a reusable plugin, not just a benchmark.
Original Excerpt
When DeepSeek released the R1 reasoning model, the breakthrough wasn’t a benchmark number — it was that the chain-of-thought was published… Continue reading on Medium »