Recently, several AI projects that I have been paying attention to have made progress. Nillion has successfully entered the Binance Launch Pool, and Mira will soon launch a public testing network as well. When using AI, everyone has experienced the illusion of AI. AI confidently presented you with a research paper that appeared to have a nose, eyes, and a source. However, when you eagerly opened the paper, you found that everything from the results to the source had been fabricated by AI. This cannot be entirely blamed on AI, because current generative AI essentially relies on "prediction" to generate each word, so based on this principle, it naturally produces illusions. So essentially, you can't use AI as a search engine. To address this issue, both DeepSeek and Grok3 subsequently launched large-scale search functions, which reduced the probability of some illusions due to the increase in references. But everyone wants lower - they just want to use AI as a search engine. Therefore, Mira has also proposed its own solution, which is to use something generated by one AI, and then arrange 2-5 AI to review it for you. For AI, generation and review are two different tasks, which is a bit like the "peer review" you conduct when writing academic papers, hoping to reduce the probability of hallucinations through this approach. According to the test data, if the AI accuracy rate is around 73% once, it can reach 96% after passing Mira's AI peer review three times, which is a good score. The main idea is' listening to both sides leads to understanding ', so this solution is a good idea, which is why it has been favored by Framework and Mechanism.