Analyzing the distributed AI computing network Gensyn

On June 12, Gensyn, a UK-based blockchain AI computing protocol, announced the completion of a $43 million Series A round led by a16z. In this revolution in the field of AI, Gensyn took the lead in handing in an answer sheet for us.

**Gensyn is essentially a first-layer proof-of-stake blockchain based on the Substripe protocol, which can promote machine learning task allocation and rewards through smart contracts to quickly realize the learning ability of AI models and reduce the cost of deep learning training. price. **The cost of a single training session of GPT-3 in 2020 is ~$12 million, which is more than 270 times higher than the estimated value of ~$43,000 for GPT-2 training in 2019. In general, the model complexity (size) of the best neural networks currently doubles every three months. The hourly cost of Gensyn's machine learning training work is about $0.4, which is far lower than the required costs of AWS ($2) and GCP ($2.5). Gensyn wants to use blockchain and other technologies to implement a decentralized large-scale distributed deep learning efficient computing protocol, and has a probabilistic learning proof and a cryptocurrency incentive mechanism.

**Gensyn connects a developer (anyone who can train a machine learning model) with a solver (a Solver, anyone who wants to train a machine learning model on their own machine). **By leveraging idle ML-capable computing devices around the world, such as small data centers, gaming PCs, M1 and M2 Macs, and even smartphones, and connecting them into a global ML supercluster, the power of ML 10-100x increase in available computing power. At the same time, Gensyn uses an innovative verification system and computing power supply to achieve ultra-large-scale and low-cost training of neural networks without trust.

1. Innovative verification system

**The core challenge of Gensyn is to verify that the computing tasks performed on the device have been performed correctly and trigger payment through tokens. **Gensyn system mainly solves the verification problem through three concepts, including probability proof-of-learning, graph-based precise positioning protocol and Truebit-style incentive game.

Consists of four main actors, including committers, solvers, verifiers, and whistleblowers. Submitters are system end users who provide tasks to be computed and pay for completed units of work. The solver is the main working part of the system, performing model training and generating proofs for verification by the verifier. The verifier links a non-deterministic training procedure to a deterministic linear computation, replicating part of the solver's proof, and comparing the distance to an expected threshold. Whistleblowers are the last line of defense, checking the work of validators and challenging them for jackpots.

The system can do all of this without trust and with an overhead that scales linearly with model size, keeping verification costs constant. The innovation of the system lies in the combination of model training checkpoints and probability checks terminated on the chain, which effectively solves the state dependence problem in neural network training of any scale.

2. New computing power supply

**Gensyn systems utilize underutilized and non-optimized computing device resources. **These devices range from currently unused gaming GPUs to GPU miners from the pre-Ethereum PoW era. Because the protocol is decentralized, meaning it will ultimately be governed by a community majority and cannot be "shut down" without community consent, unlike web2, this makes it censorship resistant. The innovation of this agreement is to make full use of unused computing equipment resources, provide more computing power for the community, and also provide a new source of income for those who own unused equipment. And the Gensyn protocol offers a similar cost to owning a GPU in a data center, and it can scale beyond AWS.

In short, **Gensyn's core goal is to democratize AI through a decentralized plan, so that more people can participate in the innovation and application of AI technology. **The core idea of this program is to use underutilized computing equipment resources to improve the efficiency and accuracy of AI models by building an open and decentralized verification system, and to provide more opportunities for AI entrepreneurs and possibility. It is an innovative and forward-looking program that is expected to play an important role in the field of AI in the future.