A few days ago, I saw that @ MiraN_Network posted a tweet with the following content: 📷 soon， Many people may think they have already taken a snapshot. But from a literal perspective, SOON means' soon 'or' soon '. So it's clear that a snapshot is about to be taken, and I've been working on Camp lately and haven't come to study Mira. Today, I also came to roll Mira up with a last-minute mentality. If there is any, then there is none 🤣 What if we catch the last bus? During the period when we know about AI and are proficient in using it, the biggest trouble that AI brings is the difficulty in distinguishing between true and false. At the beginning, GPT can execute programs to write code, and simple code logic runs smoothly on the framework without any problems. Let me give an example myself. While experiencing the convenience brought by AI in depth, I wanted to create an offline translation version of the translation machine. However, the URL corresponding to the word library he provided me with was indeed open source. But due to the passage of time, the open source word library could no longer be used, and the publicly available API interface information at that time was no longer valid. Causing the entire code to not run, I thought it might take 20 minutes to simply build it in Python, but it took me a long time to get it done. As the AI searched for more and more information through public data, it became increasingly difficult to use in practice. After studying @ MiraN_Network, it was found that what they need to do now is to fundamentally solve this problem. They launched Mira Verify, a system that uses "multi model consensus" to identify AI illusions and false information. To put it bluntly, Mira wants to be the "fact signing" of the AI era. When we encounter information sources that are difficult to distinguish between true and false, we usually rely on manual verification to verify them. To be honest, this method is too inefficient. And now AI has become an accelerator for creating illusions and errors. The big model comes as soon as it opens its mouth, and the authenticity of the content is extremely low, but the response it gives you is serious. During use, I discovered frequent errors and then fed them back to the AI. The AI provided you with the output of how to solve this error, only to realize that it had been wrong since the previous step. The whole process was quite torturous. Mira's solution is very interesting. Instead of letting one model the final say, let "a group of models" independently verify each information point. How did it do it? Its logic is to first extract all 'factual statements' from a piece of content - whether it is written by humans or generated by AI - and then broadcast these statements to nodes across the entire network. Each node runs a different model to evaluate these statements and determine whether their output is "true" or "false". Mira aggregates these judgments through consensus algorithms: If all models say 'true', then mark it as ✅ If all models say 'false', mark it as ❌ If there is a disagreement, it will be displayed as' no consensus' The content that end users see is the annotated version - which content is trustworthy, which is fake, and which is still controversial. It's like putting a 'version control label of truth' on the content. The core logic of the entire system operation process is the construction of the "truth computing layer" one ️⃣ This mechanism is de platformized. It is not who says the final word, but the systematic consensus of the model determines the authenticity. two ️⃣ It is programmable. Based on this mechanism, a front-end can be built that only displays verified true content. three ️⃣ It has the potential for economic incentives. In the future, nodes that participate in verification and provide computing resources may receive token incentives. In fact, if we follow the big logic that time is money, the purpose of using AI is actually to "reduce costs and increase efficiency". After all, everyone is quite busy and really doesn't have time to do manual verification, and Mira's consensus review solution can solve the above problems. At present, the project is still in its early stages, but a waiting list has been opened. Interested parties can fill in their email on the official website to apply for early use.