EigenCloud Trust Models.
The Vision of EigenCloud is to let developers choose the trust model they want for the offchain computation.
Between the cloud and the onchain extremes there is a range of options each with a different tradeoff on programmability.
Liveness.
In any of these scenarios, we need liveness guarantees. To obtain that the node running EigenCompute or EigenAI needs to be staked on EigenLayer and write data to EigenDA.
If the node goes offline, the node is slashed, a new node can stake, download the data from EigenDA and restart the computation. Thus the computation remains unstoppable.
ZK for safety.
Zk is the strongest mechanism for safety but requires writing a prover as well as paying the prover cost. The prover cost can be as much as 100,000 times the cost of direct computation, Thus it remains infeasible for running a MMORPG game, a twitter recommendation algorithm, or running the best AI agent.
Cryptoeconomic safety.
To get objective cryptoeconomic safety, we need a) deterministic re-execution, b) onchain DA and c) onchain fraud proofs. If any of these conditions are not satisfied, we do not get full objective cryptoeconomics.
To get intersubjective safety from EIGEN staked, we need a) deterministic re-execution and b) EigenDA for DA.
TEE safety.
TEE, under its trust model, can guarantee correct program execution, including correct data imports and private key storage. No other mode can guarantee correctness for a) non-deterministic computation, b) time-varying data imports and c) private computation.
Mainnet Alpha launch principles.
We launched EigenCompute and EigenAI on Mainnet Alpha. The goal of the alpha launch is to provide developers early permissioned access to these powerful features so we can prioritize the features based on developer feedback. In the tradeoff between programmability and verifiability, we are starting with the most programmable version and then gradually increasing the level of trust and verifiability. Both these products will need to go through thorough audits before opening up for permissionless access.
EigenCompute
Because one of the main usecases is to launch verifiable AI agents, and such agents need features like deep search - which are time-varying and cannot be handled on any mode other than TEE, we are launching with the TEE mode for EigenCompute. The EigenLayer ecosystem has very powerful coprocessors with deterministic verifiability already including Layer WAVS and Brevis (which includes both cryptoeconomic and ZK proofs). We will be expanding EigenCompute to other modes soon.
EigenAI
To have inference run on the most powerful open source models, we need the ability to run AI inference on multiple GPUs. The first major result for EigenAI is a new stack that can get deterministic re-execution on multi-GPU execution without any backoff in speed of inference. This determinism requires a fixed GPU type (like H100) to re-execute. We note that running inference on deterministic CPU (like Cartesi VM or RiscV) is possible but will require a massive slowdown making it impossible to run state-of-the-art models. Hence, in the spirit of maximizing programmability, we use cryptoeconomic security. While the initial model will require intersubjective verification, it is possible to write onchain fraud proofs to create fully deterministic models for re-execution. To get full onchain security, we note that the inference results may need to be written to an onchain DA too.
Roadmap
Over time, we will fill in the entire graph between programmability and verifiability for both EigenCompute and EigenAI so that developers can get full choice for any offchain computation. EigenLayer has a thriving ecosystem of partners who are building these different modalities and we will be relying on the ecosytem to offer these multiple options in a short period of time.
19.5K
104
The content on this page is provided by third parties. Unless otherwise stated, OKX is not the author of the cited article(s) and does not claim any copyright in the materials. The content is provided for informational purposes only and does not represent the views of OKX. It is not intended to be an endorsement of any kind and should not be considered investment advice or a solicitation to buy or sell digital assets. To the extent generative AI is utilized to provide summaries or other information, such AI generated content may be inaccurate or inconsistent. Please read the linked article for more details and information. OKX is not responsible for content hosted on third party sites. Digital asset holdings, including stablecoins and NFTs, involve a high degree of risk and can fluctuate greatly. You should carefully consider whether trading or holding digital assets is suitable for you in light of your financial condition.