Someone in subnet number 96 said FLock OFF The science fiction novel "Ready Player One" by American author Brian is very good. It describes the protagonist Wade's random shuttle between various subnets in the virtual world "Oasis" and a series of magical adventure stories. I have also been caught up in a network recently, unable to extricate myself. It's not a love affair, but a subnet with the number SN96. Let's sneak in! ⬇️ I don't like both 6 and 9, but when combined, I really like them. So when I first saw SN96, I became addicted. SN96 is the 96th subnet on the Bittensor network, also known as FLock OFF @ floor_off_stn96. 96, F * * * OFF This is too unrestrained, it's my favorite style. After careful study, it is indeed well deserved: SN96 advocates open motivation, federated learning, fearlessness of authority, and says F * * * OFF against outdated centralized control and privacy breaches. This is a soft core science popularization article, through which you can have a partial understanding of the following knowledge: What is SN96 What is Federated Learning ▰ What is edge computing Who is an undercover agent and how to cheat in the game I know it's a bit too broad, but I will try my best to be easy to understand and humorous, and try to be as short, shallow, and slow as possible one ️⃣ The sudden rise of SN96 The TAO community has been a bit too lively lately. There is a new subnet online called FLock OFF, with the number 96, which has caused a stir in the entire community. Even the founder of Bittensor, @ conts_born, personally came out to watch, which is not a joke. What does' Flock OFF 'mean? It may sound a bit mischievous, but there is actually a deeper meaning behind it. The name FLock OFF itself carries a "free from constraints" attitude, calling on everyone not to be kidnapped by centralized AI models anymore. But to be more serious, it's actually a combination of some words: FLock: A privacy training platform created by http://(((((FLock. io)))))) Open: Participation without permission Federated: Federated Learning Framework: Decentralized AI Training and Validation Framework The goal of FLock OFF is to create an open-source framework specifically designed for building high-quality small language model training datasets, including supervised fine-tuning (SFT) and direct preference optimization (DPO) data, to help improve model performance. Federated learning enables efficient training of small language models on devices such as smartphones, making it particularly suitable for scenarios with limited network and computing power. This method allows small models to continuously improve on edge devices, and the final performance may even exceed that of simplified large models such as GPT-4-nano. In less than a month, FLock OFF's FDV reached 310 million US dollars, with 780 holders, 251 active miners, and 5 validators, which is definitely the most impressive data. Why is it so powerful? It's simple, because it was the super strong regular army that created SN96 that brought about the actual landing. two ️⃣ The regular army behind SN 96 What the regular army is talking about is developing @ floor_io for SN96, which is a privacy model training platform that shouts' Not your models, not your AI '. In order for the training nodes of federated learning to work efficiently, each device only requires a small but high-quality amount of data. SN96 technology can intelligently filter higher quality data and study how to compress data while ensuring that the trained model still maintains strong performance. Let's sort out the business process between http://(((((((FLock. io))))))) and SN96. As a subnet on the Bittensor network, SN96 incentivizes miners to upload high-quality data through its decentralized dataset quality competition network, while validators use LoRA training to evaluate data quality and ensure only the best dataset is selected. This process is crucial for providing high-quality and diverse data for the next stage. On the other hand, http://(((((((((FLock. io)))))))) promotes federated fine-tuning of these datasets in FL Alliance, a collaborative network that utilizes federated learning to protect data privacy while optimizing model performance. The role of http://((((((((FLock. io)))))) is to coordinate this fine-tuning process and ensure efficient training of the model on distributed nodes. Finally, the trained model is deployed through Moonbase, which is http://((((((((((FLock. io))))))))))))))))))), a practical infrastructure that supports real-world applications, completing the entire cycle from training to deployment. FLock OFF is the research base of the team at Bittensor and a supplement to the FLock ecosystem. As the team behind SN96, http://(((((((FLock. io))))) has partnered with mainstream AI teams such as Alibaba's Qianwen Qwen to become the first decentralized AI training platform integrated by a leading open-source model provider. This is not a joke, indicating that the technology of http://(((((((FLock. io))))) has been recognized by the mainstream AI community, and both parties may expand their landing space in the field of larger scale model collaborative training and deployment. The model of http://(((((((FLock. io)))))) has been successfully applied in multiple projects, particularly in the fields of government and finance, by ensuring data privacy and efficient computation through its decentralized collaboration mechanism. For example, in collaboration with Animoca Brands, http://((((((((FLock. io))))))) assisted in the development of (((((((((FLock. io))))))))) http://heyani.xyz The project utilizes its AI model to enhance the due diligence, investment, and market production capabilities of blockchain applications, ensuring data privacy and efficient collaboration. In the cooperation with UCL Hospital, through its overall solution of privacy training, it has provided hospitals and patients with a complete blood glucose monitoring and prediction model to protect the life and health of thousands of blood diabetes patients. Oh, by the way, http://((((((((FLock. io)))))))) also supports Base's MCP project, helping it to implement AI applications and optimize the efficiency of decentralized authentication and smart contract execution. three ️⃣ What can SN96 do In summary, the creation, validation, and motivation of decentralized datasets. There are several examples in the book 'Out of Control' that left a deep impression on me. Bees can build geometrically perfect beehives without unified leadership deployment, while ants can complete large-scale cargo logistics transportation without leadership supervision. The decentralization or lack of centralization in nature provides the opposite answer to what human organizations have always believed requires centralized decision-making to accomplish. FLock OFF is a decentralized system where miners compete by uploading high-quality datasets aimed at improving the performance of AI models. The verifier evaluates the quality of the dataset through LoRA (Low Rank Adaptation) training. These are not achieved through unified deployment and supervision, but through invisible hands. Miners receive token rewards for providing high-quality datasets, while validators receive token rewards for accurately evaluating the quality of datasets. This mechanism optimizes the contribution quality of participants through a game theory framework. Through Bittensor's subnet architecture, FLock OFF allows global participants to contribute data and computing power in a decentralized manner, avoiding centralized entities from controlling data. This is FLock OFF's emphasis on "Incentives driven trustless coordination", which ensures that participants contribute valuable data through economic incentives while maintaining the fairness and decentralization of the system. four ️⃣ What is Federated Learning I can't take it anymore, I've mentioned this word several times, I have to figure it out. Have you ever played an undercover game? If you haven't played before, you can check out this episode of @ degirls. The rules of the undercover game are very simple. Without sharing raw data (identity), everyone trains a mental model (guess) by asking and answering questions to find out who the undercover agent is. How does federated learning train AI? Specifically, everyone is local and uses local devices to train the model. Only the updated information of the model is uploaded to the central server, which aggregates these updates to improve the overall model. This model training method, combined with the low requirements of edge computing on devices, allows Flock OFF to select the most valuable samples for model performance improvement from a large number of data tasks. It means that it does not participate in model training and deployment, but strictly screens and controls the data, which directly determines the final performance of the model. Oh, by the way, how to cheat in Who is an undercover game? It's actually the drawbacks of centralization, such as premature identity leakage, signal transmission, manipulation of discussions, external assistance, and so on. Using the token economy system to maintain network security and efficiency: Economic incentives and punishments: Participants need to pledge tokens in advance to join the network. If a node commits wrongdoing (such as submitting false data), the system will slash its pledged tokens, increase the cost of wrongdoing, and thus reduce bad behavior. Blockchain traceability and tamper proof: All data operations will be recorded on the blockchain to ensure traceability. The false information submitted by malicious nodes will be identified and removed to ensure that the final trained model is of high quality and reliable. // In this decentralized AI market, everyone is a participant and a builder. The addition of http://((((((((((FLock. io))))))) has made SN96 more exciting, as evidenced by the lively community response. I hope this article gives you a deeper understanding of http://(((((((((((FLock. io))))), SN96, Bittensor subnet, federated learning, and edge computing. After all, the world of AI is not just about GPT-4, there are countless possibilities waiting for us to explore. so, FLock OFF, Get's Go!