MindLayer: how we help AI networks
Current Restaking Challenges
Restaking leverages Ethereum’s consensus layer to extend cryptoeconomic security to additional applications on the network. Ethereum is the most secure PoS chain in the world with $50B in total value locked. But the average staking yield is ~4%. Bitcoin’s current market cap is currently at $1.4T and obviously, staking is not an option. Projects like Babylon are using a self custody method and locking Bitcoin to secure proof of stake chains.
However, challenges are associated with restaking:We identify two challenges in current restaking design:
Centralization Risk: one major concern is that restaking could result in stake centralization, leading to a few validators cheating and controlling the stake. This could lead to a potential loss of neutrality.
Compounded Risk: LSTs from one ecosystem (e.g. ETH) may suffer and amplify the compounded risk to become systematic for the ecosystem.
Data Security: operators are not equipped with data protection mechanisms, which may impact data privacy and computation fairness. It is critical for data intensive networks like AI or DePin networks.
When Restake meet AI
AI projects and systems are growing exponentially. AI compute demand is doubling every 3.5 months. However, it has challenges with highly centralized GPU resources and governance model to develop the main AI development, e.g. model training, inference serving. It has already faced the bottlenet, not only to the huge amount of cost overhead, but also to ethics and privacy. We, as Mind Network, are prepared for AI adoption with blockchain with three enhancement contributions for blockchain and AI together:
Enhancement 0: the security of crypto economy is critical and especially for cold-start AI networks. We introduce diversified restaking tokens to eliminate the compounded risk from single ecosystem.
Enhancement 1: we believed the demand and potential for AI networks are massive, and this will allow restaking token holders to share the upside of bullish AI tokens with minimized volatility risks.
Enhancement 2: AI models need huge amounts of parameters to be collaboratively updated. We introduce an new AI native rollup with faster computation and lower cost, and also an innovative Proof of Intelligence consensus designed for AI networks.
Enhancement 3: If AI parameters are purely in plain data and every operators can see it. It can easily course copy from each other to cause manipulation. With the protection from FHE, we enable anonymous scoring and voting to keep AI secret, but can be publicly verifiable.
Last updated
