Will decentralized protocols be the future of machine learning infrastructure?
Increased interest in artificial intelligence (AI) and machine learning (ML) has led to a shortage of hardware resources and inflated cloud service costs, but decentralized infrastructure can challenge the reliance on centralized players.
Harry Grieve, co-founder of machine learning computing network Gensyn, spoke exclusively to Cointelegraph during the ETHGlobal event in London about the promise of peer-to-peer computing networks that will challenge Web2 services like Amazon Web Services.
Gensyn is a decentralized network that allows people to connect with various tools on the Internet to train machine learning models. The company is backed by several Web3 venture capital firms and has raised $50 million from Andreessen Horowitz in 2023.
Grieve says the network has great potential as the Internet evolves into a dynamic representation of information that enhances “self-sovereignty and the freedom to compute online.”
Gensyn, along with Grieve and co-founder Ben Fielding, has been developing machine learning computing for training and decentralized verifiable systems since 2020. The pair have been looking to solve the triple dilemma with blockchain-based technology.
“How do you equate to another device and train a machine learning model on that device, if the device is unreliable? B), your training model can't scale on that single device. And C, you want a measure of the system's total and unit economic results that is as good as AWS can achieve.” ” Grieve said.
Jensen's paper describes the protocol as a “layer-1 trustless protocol for deep learning computing.” The network directly and instantaneously rewards participants for using the network's computing resources and performing ML tasks.
Grieve says the challenge of building the network lies in validating the work of ML, which sits at the intersection of complexity theory, game theory, cryptography and optimization.
“We realized that if you're going to build this in a decentralized way, you need a way to reach decentralized consensus about who did what. That's basically what blockchain is all about.”
Taking a leaf out of Satoshi's Bitcoin playbook
Gensyn takes some inspiration from the concepts behind the Bitcoin protocol, and Grieve says he's a big believer in the old days of Bitcoin laptop mining, when it was still possible to use smaller devices where users could earn BTC.
“Satoshi has re-enfranchised the money people have; they can generate it from a laptop. They can turn electricity into money. They can turn fiat into something hard.”
The long-term plan is to make Gensyn a tool that allows a wider set of users and hardware to provide or access computing resources for ML training, but the initial launch will target users with more graphics processing units (GPUs) because they represent the fastest. Way to get more feedback.
We are actively planning for people to build on Jensin to make user-friendly, delivery and contribution applications. Finally, an individual with a laptop can download our client and run it in a way that connects you to the network,” Grieve said.
Related: io.net Adds Apple Chip Support to Cloud-Based GPU Source Network
Apple's silicon chips promise to unlock enormous global computing resources. Grieve's research on Apple's M2 and M3 chips shows that the hardware is on par with mid-range current-generation consumer Nvidia RTX GPUs.
This provides two major advantages for protocols such as Gensyn, which can tap into a wider set of devices to contribute to a global supercluster. Sadness says:
“I think people are more likely to have something that looks like a MacBook than a discrete GPU, so if you can contribute that, it's more of a democratization.”
Apple's silicon chips are also very versatile as a “system on a chip” that can be replicated by other chip manufacturers.
Grieve believes that future smartphones will likely have more edge devices that are more powerful, saying that decentralization and device-agnostic authentication across multiple devices is important.
As previously reported by Cointelegraph, Solana-based decentralized network io.net for artificial intelligence and machine learning services will be based on Apple's silicon chip hardware.
Magazine: Is Measuring Blockchain Transactions Per Second (TPS) Foolish in 2024? Big questions