First, let's take a look at Mira's development documentation. The main core features of the SDK that have been released are not only initiating tasks for large language models, but more importantly, "intelligent routing" and "load balancing", which automatically select the most suitable model or can put pressure on multiple models when the task is difficult. Then I flipped through their white paper, which actually aimed to address the verifiability of AI output results, that is, how to consistently produce reliable, verifiable, and almost error free results, reducing AI's illusions and biases. The process is similar to this: Disassemble verification targets → Distribute to multiple model nodes → Use blockchain and economic models to promote node honesty → Finally output consensus verified results However, based on the current testing network and development documentation, the amount of external work is still very simple, and only tasks for some mainstream language models can be initiated. It is unclear how these "intelligent routing" and "load balancing" will be implemented. We look forward to gradually achieving the goal of "making AI reliable by verifying outputs" in the future.