By spreading the control across a decentralized network, it becomes much harder for any external force to censor or manipulate the information as the model is 100% transparent, ensuring that users receive unaltered and truthful data.

Perception management is a tale as old as time, from the days of Rome by using the gladiator fights as a tool to diverting attention from societal issues, to modern days with the main stream media.

Decentralized AI uses the same peer-to-peer logic as blockchains like Ethereum or Solana do. Instead of a model being hosted by one entity, they are distributed globally between multiple nodes all validating each other in real time.

If any one node goes offline, the model isn't disrupted as long as there are other nodes online.

Models are split between multiple nodes, each hosting specific transformer blocks of the model. Each server holds several consecutive blocks, the number of which depends on the server’s available GPU memory.

Subnets are inherently fault-tolerant because they distribute data, control, and operations across multiple nodes or participants, rather than relying on a single centralized server or point of control. This design ensures that even if some components of the system fail or are compromised, the overall system can continue functioning without interruption. As long as all of the subnet AI model layers are being served by subnet nodes, the subnet will continue to operate in a healthy state.

Subnet validator nodes can use the Hypertensor blockchain to form a consensus using the SCP (Subnet Consensus Protocol) that includes an incentives mechanism, a fault-proofs mechanism, and an accounting mechanism.

The DSN (Decentralized Subnet) standard uses a proof-of-stake mechanism as one of its consensus mechanisms. It works in the same way a blockchain works; any node entering the network is verified to be staked on-chain before they enter other nodes routing tables. After entry, any messaging between peers, such as running inference sequences, must also be validated on both incoming and outgoing communications.

When generating tokens, a client stores the model’s token embeddings (which typically comprise a small fraction of the total parameter count and can fit in RAM in most modern laptops, servers, and workstations) locally and relies on servers to run Transformer blocks. Before each inference session, the client finds a chain of servers that collectively hold all model layers. Once the chain is formed, the client uses the local embedding layer to look up embedding vectors for prefix tokens, then sends those vectors to servers and receives new representations. Once the client obtains the outputs of the final block, it computes the next token probabilities and repeats this process. While the session is active, servers store attention keys and values from past client inputs and use them for subsequent inference steps. Clients also store past inputs to each server so that if any server fails or goes offline, another one can quickly take its place.

Currently, Ethereum has over one million validators ensuring the security and liveliness of the Ethereum blockchain. Soon, decentralized AI will have millions of validators serving and validating AI models for the world across thousands of models.