Next-Gen AI Computation: CamelidCoin’s Blockchain Solution
Read the whitepaper: https://camelidcoin.org/
This video in blog form: http://dylandunn.me/post/camelidcoin-...
CamelidCoin is a trustless, incentive-based blockchain protocol de-
signed for distributed auto-regressive large language model (LLM) com-
putation and training. The system is built on a peer-to-peer network
where lightweight clients can submit input for completion by a large pool
of compute nodes. In exchange for their work, the compute nodes are com-
pensated by the client. To ensure the validity of the output from these
compute nodes, we propose a new algorithm called (Random Autoregres-
sive Subsampling for Token Integrity Checking) (RASTiC), which can be
completed either by the client or other compute nodes. This algorithm
quickly verifies the output’s authenticity with near certainty. Addition-
ally, throughout the video, we will detail the protocol’s structure and
specifics. Our protocol works off of the original peer-to-peer electronic
cash system proposed by Satoshi Nakamoto’s Bitcoin and modifies it to
fit our use case. We will also discuss current challenges and limitations
with our current protocol.