AGI

AGI model data consists of 4 components (model code, model parameters, training data = validation data), with each part iterating independently. Except for the model code, the other components have enormous data volumes, so they need to be stored in shards.
Each data component (model parameters, training data, validation data) is processed through error correction codes and split into k shares, distributed redundantly across n relays. Probabilistically, users only need to randomly find m relays from all n relays to fully reconstruct the complete data. (n>m>k)
- For users' private AGI models, they are first encrypted, then processed through error correction codes and split into k shares, distributed redundantly across n relays.
Data is primarily divided into two categories: Foundation models shared by all humanity and private Domain-specific models (DSM).
- Foundation models (massive) are stored in shards across all relays in the network
- DSMs (smaller) are stored either in shards or without sharding in a subset of relays
Private domain-specific models are smaller because they specify a particular version/current version of the Foundation model as master, then the private model acts as its fork, storing only the delta between the private model and Foundation model as $\Delta = dsm-fm$, resulting in a compressible (low-rank) matrix. [1]

Nostr

Nostr has two components: clients and relays. Each user runs a client, and anyone can run a relay.
Each user is identified by a public key, every piece of data has a signature, and each client verifies these signatures.
At startup, the client connects to all m relays it knows about.
Clients fetch their subscribed data (Foundation model and/or domain-specific model) from m relays they choose and trust.
They publish their private DSM data, fine-tuned from the Foundation model data, to other relays of their choice.
Relays do not communicate with other relays, only directly with users.

Competition models are not put on-chain, only championship models from each round are "on-chain".

Training data is provided through DAO contributions, with no incentives other than A Fully Decentralized Crowdsourced Valuation Protocol for Non-depreciable Assets: P.I.V.O.T.
Anyone who can advance the evolution of the Foundation model (compression rate) will receive PoWoC mining reward tokens, which then serve as PoS mining and become the main source of block generation.

Quantification Method of PoWoC
The DAO will airdrop many SBT governance tokens to respected scholars across various fields.