Current Inefficiencies
Till date, Bittensor is the only project that we consider has meaningfully attempted to design an incentive layer for decentralized intelligence but we view the following as inefficiencies in design:
High cost of entry: Requiring at least 100 TAO to start a subnet prices out majority of innovators who wish to build intelligence on-chain. The perception of "competition" does more harm than benefit the DeAI ecosystem. The cost to setup a company off-chain is closer to 1 TAO at current prices in most jurisdictions around the world.
Double spend of compute + centralization of validation: In the Bittensor Network, subnet validators give subnet miners some work to do while simultaneously giving the same work task to others in the subnet miners. All subnet miners complete the task and respond to the subnet validator with the work results.
The subnet validator then ranks the quality of the work done by the subnet miners. Source
Validators defining real or useful work is a recipe for disaster. Our team spent many years working on Filecoin Plus, a similar design that led to rampant block reward abuse within the Filecoin Ecosystem and eventually needed to be decentralized and automated. Such a design drastically increases abuse in the system, eventually leading to a winner takes all scenario.
Ejection without cause: Dethroning subnets for lack of "usefulness", similar to Polkadot parachains, is an inefficient mechanism as it deprives builders the opportunity to iterate and improve on their offering to build a moat for their IP and progressively monetize the AI assets over a longer time horizon.
Last updated