Interview with Gensyn co-founder Ben Fielding: How can the decentralized computing protocol led by a16z achieve the democratization of AI?

Interview: Sunny and Min, coderworld

Guest: Ben Fielding, Co-Founder of Gensyn

Our goal is not to monopolize the entire machine learning ecosystem, but to establish Gensyn as a protocol that optimizes the utilization of computing resources, just above electricity, to significantly improve humanity's ability to effectively use computing resources.

—Ben Fielding, Gensyn CoFounder

January 2024, OpenAI CEO Sam said,The two important "currencies" of the future will beComputing power and energy.

However, asAIThe power currency of the era, computing power is often monopolized by large companies, especially in the field of AGI large models. Since there is monopoly, there is also the power of anti-monopoly. Decentralized artificial intelligence (Decentralized AI) AI) came into being.

“BlockchainThis permissionless component can create a market for buyers and sellers of computing power (or any other type of digital resource, such as data or algorithms) to trade globally without middlemen," said a16z, a well-known investment institution.An articleHow to explain AI computing powerBlockchainThe description item is Gensyn.

Gensyn is a decentralized deep learning computing protocol.Designed to be the foundational layer for machine learning computing,contractThis approach can promote task allocation and rewards for machine learning, quickly realize the learning ability of AI models, and reduce the price of deep learning training.

Gensyn connects developers (anyone who can train machine learning models) with solvers (anyone who wants to train machine learning models on their own machines), increasing the available computing power for machine learning by 10-100 times by leveraging the long tail of idle machine learning-capable computing devices (e.g., small data centers, personal gaming PCs) around the world.

In summary, Gensyn's core goal is toBlockchainPlan to democratize AI.

In June 2023, Gensyn announced the completion of a US$43 million Series A financing round, led by a16z, with participation from CoinFund, Canonical Crypto, Protocol Labs, Eden Block and others.

Gensyn was founded in 2020 by computer science and machine learning research veterans Ben Fielding and Harry Grieve. Harry Grieve studied at Brown University and Aberdeen University and is a data scientist and entrepreneur; Ben Fielding graduated from Northumbria University and served as co-founder of SaaS platform Fair Custodian and director of Research Analytics.

Coderworld interviewed Gensyn co-founder Ben Fielding to learn about his crypto AI journey and Gensyn’s AI weapons.

Gensyn’s value proposition from the founder’s perspective

coderworld: What inspired you to start Gensyn?

Ben:

My original background is in academia, as a machine learning researcher focusing on the field of Neural Architecture Search, which involves optimizing the structure of deep neural networks, especially for computer vision applications.

My work is developing algorithms to evolve neural network architectures in a swarm manner. This process involves training many candidate model architectures simultaneously and gradually evolving them into a single meta-model optimized for a specific task.

During this time, I encountered significant challenges related to computing resources. As a PhD student, I had access to several high-performance GPUs, which sat under my desk in a large workstation that I had managed to purchase.

Meanwhile, companies like Google are doing similar research, but they’re using thousands of GPUs and TPUs in data centers, running for weeks at a time.This disparity made me realize that despite having all the necessary resources except ample computing power, others around the world face the same limitations, which hinder the rate of research and social progress.I was unhappy with the situation, which was ultimately why we created Gensyn.

Prior to working full-time at Gensyn, I spent two years co-founding a data privacy startup focused on managing consumer data flows and consent-based access to user data, with the goal of improving how individuals and businesses interact with data.

The experience taught me valuable lessons, including common entrepreneurial pitfalls, and reinforced my cautious approach to personal data flows and consent-based access.

Four years ago, I closed my startup and joined Entrepreneur First, an accelerator in London, where I met my partner Harry Grieve. It was there that we started Gensyn with the goal of solving the challenge of global computing resources. Our initial strategy involved distributing computing tasks (federated learning) across private data silos of a single organization, which was very interesting. We quickly realized the broader potential of expanding this approach to the world.To address this expanded vision, we must address fundamental trust issues related to the source of computing itself.

Gensyn has since been working to ensure the accuracy of machine learning tasks processed on-device through a combination of proofs, game-theoretic incentives, and probabilistic checks. While the specifics can get pretty technical, Gensyn is working to develop a system that enables anyone in the world, using any computing device, to train machine learning models.

coderworld: Sam Altman needs $7 trillion to run AI chip factories to address the global chip shortage. Is his plan realistic in terms of scaling chip supply? In the meantime, what are Gensyn addressing?Xiaobai NavigationDifferent from Altman's solutionAIquestion?

Ben:

Regarding the field of AI and the challenges it faces, Gensyn is solving problems similar to those faced by Altman. Essentially, there are two ways to solve the problem of access to compute. Machine learning is becoming more and more pervasive and will likely be integrated into every technology we use, transitioning from imperative code to probabilistic models. These models require a lot of computing power.When you compare the demand for computing to the world's chip-making capacity, you'll notice a significant gap; demand is skyrocketing, while chip production is increasing only gradually.

The solution lies in (1) making more chips to meet demand or (2) using existing chips more efficiently.

Ultimately, both strategies are necessary to address the ever-increasing demands for computing resources.

I think Altman tackled this issue head on effectively. The problem is the chip supply chain, which is a very complex system. Some parts of this supply chain are particularly challenging, and only a few companies are equipped to manage these complexities. Now, many governments are beginning to view this as a geopolitical issue, investing in domestically relocating semiconductor manufacturing plants and addressing some of the bottlenecks in the supply chain.

In my opinion, what Altman is proposing is to use a $7 trillion figure to test the market to measure the degree of concern that global financial markets have about this issue.The fact that this staggering number was not rejected outright is remarkable. It prompts people to rethink: “That sounds absurd, but is it really true?”

This response suggests that there is indeed significant concern and that people are willing to allocate significant funds to solve the problem. By setting such a high benchmark, Altman is actually creating a reference point for any future chip production efforts. This strategic move shows that even if the actual cost does not reach $7 trillion, it sets a precedent for large-scale investment in this area, demonstrating a strong commitment to solving chip manufacturing challenges.

Gensyn’s approach is different; we aim tooptimizationworldwideUse of existing chips. Many of these chips, from gaming GPUs to Macbooks equipped with M1, M2, and M3 chips, are underutilized.

These devices are fully capable of supporting AI processing without the need to develop new dedicated AI processors.Leveraging these existing resources requires a protocol that integrates them into a unified network, similar to how TCP/IP facilitates Internet communications.

Such a protocol would enable these devices to be used for computing tasks on demand.

The main difference between our protocol and traditional open source protocols such as TCP/IP is financial.While the latter is a purely technical solution, using hardware resources itself involves inherent costs, such as electricity costs and the physical cost of the hardware itself.

To address this issue, our protocol incorporatescryptocurrencyand decentralization principles to build a value coordination network to incentivize contributions from hardware owners.

Gensyn thus represents a solution of a dual nature: it is both an open source protocol for software connectivity and a financial mechanism for resource compensation.

Furthermore, the challenges facing the machine learning market extend beyond just computing resources.

• Other factors such as data access and knowledge sharing also play a key role. Through decentralized technology, we can promote the value of these different components and promote a more integrated and efficient ecosystem.Therefore, Gensyn does not operate in isolation; we solve one part of the broader challenge, but other solutions are needed to address the remaining problems. This collaborative effort is critical to advancing the field of machine learning.

Defining Gensyn's Dual-natured Solutions

coderworld:Can you explain Gensyn in the simplest terms? A dual solution??

Ben:

Simply put,Gensyn is a peer-to-peer network based on open source software, all you need to do to get your device involved is run this software, and your device must be able to perform machine learning training tasks. This network consists of multiple nodes, each of which runs the software like your device, and they communicate directly with each other, sharing information about available hardware and tasks to be performed. The benefit of this is that there is no need for a central server, your device can interact directly with other devices, thus avoiding the need for a central server.

An important feature of Gensyn is that its communication process has no central authority. For example, if you are using a MacBook, it will connect and communicate directly with other MacBooks, exchanging information about hardware capabilities and available tasks.

One of the main challenges facing Gensyn is verifying non-deterministic computations off-chain, which areBlockchainToo big.

Our solution is to introduce aAuthentication Mechanism, allowing the deviceGenerate verifiableofCalculation ProofThis proof can be checked by other devices, ensuring the integrity of the work without revealing which parts of the task are being verified, therebypreventequipmentonlyComplete the parts of the task that may be inspected.

Our system encourages devices to participate in a cryptographic proof process or selective work rerun as solution providers and verifiers to determine the validity of completed tasks. In essence, Gensyn aims to enable interoperability between nodes, mutually verify work, and reach consensus on completed tasks. Payments for tasks are executed within this framework, leveraging the trust mechanism of blockchain.This technology ecosystem mimics the functionality of Ethereum, focusing on mutual verification between nodes to ensure the integrity of the task.

Our main goal is to achieve consensus on task completion with minimal computational effort, ensuring the integrity of the system while accommodating large-scale machine learning tasks.

In summary, Gensyn can be divided into two main parts.

• The first part is the blockchain aspect, including what I mentioned earlierState Machine. This is where shared computation between participants occurs.

• The other half of Gensyn involvesCommunications Infrastructure,Focus onHow nodes interact and processMachine Learning Tasks.

This setup allows any node to perform any computation, provided it can later verify the work on the blockchain side.

We are building a communication infrastructure covering all nodes to facilitate information sharing, model partitioning when necessary, and extensive data processing. This setup supports various model training methods such as data parallelism, model parallelism, and pipeline partitioning without requiring immediate trust coordination.

Dual Solution = State Machine + Machine Learning Task Communication

Gensyn State Machine

coderworld:How does the Gensyn chainSpecific machine learningpeer to peernetworkPlay a role?

Ben:

Initially, we assume that all participants are following theirRoleWe then turn our attention to the blockchain side, where we maintain aShared State,includeHashed transactions and operations, and hash the previous block, thus forming a complete chain.

ParticipantsBetweenconsensusyes,If the calculations in a block match and produce the same hash value, the work is considered completed correctly, allowing us to move on to the next link.

Gensyn uses a POS mechanism to reward contributions that verify block generation.

Creating a block involves (1) hashing the operations required for machine learning verification work and (2) recording the transactions that occurred within the block.

While our approach is similar to systems such as Ethereum, ourUnique contribution mainly lies in communication,in particularHow nodes manage and collaborate to process machine learning tasks.

coderworld:How is the Gensyn chain different from Ethereum? If the core infrastructure is not novel, how is the PoS chain designed to meet the specific use case of machine learning?

Ben:

The core structure of our blockchain is not new, except for oneNovel data availability layerThe significant difference is that weAbility to handle larger computing tasks, which makes our operationMore efficient than is normally possible on Ethereum.

This is forConvolution operation (convolution operations) are particularly relevant, as manyMachine LearningModelBasic components.

It is challenging to perform these operations efficiently using Solidity in the Ethereum Virtual Machine (EVM).

Gensyn chains provide more flexibility, allowing us to process these calculations more efficiently without being limited by the scope of EVM operations.

The real challenge isAchieving model generalization（Generalizability）:This means that the model canMeet the newSample timeAccurately predict its location, even if it has never been seen before, because itHave a broad enough understanding of space.

This training process requiresPlenty of computing resourcesBecause it needsrepeatedlyPass data through the model.

The task of Gensyn’s machine learning runtime isGet a graphical representation of the model, and place it in a frame so thatGenerate proof of completion for each operation as it performs computation.

There is aImportant Questions,Right nowDeterminismandReproducibility.

Ideally, in the world of mathematics, repeating an operation should produce the same result. However, in the physical world of computing hardware,Unpredictable variables may cause slight changes in calculation results.

Until now, a certain degree of randomness in machine learning has been considered acceptable or even beneficial, as it helps prevent models from overfitting and promotes better generalization.

However, for Gensyn,Generalizability and reproducibility are both important.

A change in the calculation result may result in a completely different hash value, which could cause our verification system to incorrectly mark work as incomplete, risking financial loss.responseAt this point, our runtimemake sureoperateOn each deviceAllDeterministic and reproducibleYes, this is a complex but necessary solution.

This approach is somewhat similar to usingPyTorch,TensorFloworJAXMachine learning frameworks such as . Users can define models and start training in these frameworks. We areAdapting to these frameworks and underlying libraries, such as the Compute Unified Device Architecture (CUDA) to ensure that the model executes in a repeatable and accurate manner on any device.

thisEnsures that hashing the result of an operation on one device produces the same hash on another device, highlighting the importance of this aspect of machine learning execution in our system.

Gensyn decentralizes cloud services through open source blockchain communication protocols to support decentralized machine learning

coderworld:So what about this set of blockchain communication facilities specific to machine learning networks on the Gensyn chain?

Ben:

The purpose of communication infrastructure is to facilitate intercommunication between devices.Its main function is to allow one device to verify proof of work generated by another device.

In essence, the communication between devices is usedMutual verification of workThis process needs to be done through blockchain, as blockchain is the onlyActing as a central arbitratorThe blockchain is the only source of truth in our system, without it there is no way to reliably verify the identities of the parties involved and anyone could claim they have verified work.

Blockchain and its encryption technology make identity verification and work confirmation easySafety. The device can prove its identity andSafetySubmit information so that other parties can identify and verify the authenticity of that information.

This systemThe ultimate goal is to provide compensation to device ownersIf you have hardware that can perform machine learning tasks, you can rent it out.

However,In traditional systems, this process is complex and costlyFor example, buying a large number of Nvidia GPUs and renting them out — turning capital expenditures into operating expenses, similar to cloud service providers — involves numerous challenges. You need to find AI companies interested in your hardware, build sales channels, develop infrastructure for model delivery and access, and manage legal and operational agreements including service-level agreements (SLAs). SLAs require onsite engineers to ensure uptime agreed with customers, and any downtime will result in contract-based liabilities and potential financial risks.This complexity is a significant barrier for individuals or small businesses, and is one of the reasons why centralized cloud services have become mainstream.

Gensyn provides a more effective method.Eliminates the human and business costs typically involved in these transactionsYou just run some software without relying on legal contracts and engineers to build infrastructure.Legal agreements are smartcontractInstead, work verification is performed by automated systems, checking that the task was completed correctly. No longer do you have to manually process breach of contract claims or seek legal settlements, all of this can be solved instantly by technology, which is a significant advantage. This means that suppliers can immediately get benefits from their GPUs by just running some software without any additional hassle.

About Go to Market

The way we encourage vendors to join the Gensyn network is by telling them that they can immediately enter the in-demand market for machine learning computing by running open source software. This is an unprecedented opportunity that significantly expands the market and allows new entrants tochallengepictureAWSSuch traditional servicesDominanceAWS and other companies need to manage complex operations, and we are transforming those operations into code, creating new ways for value to flow.

Traditionally, if you have a machine learning model that needs training and are willing to pay for compute, your money will go to the major cloud providers that have a monopoly on supply. They dominate the market because they can manage it efficiently. Despite growing competition from Google Cloud, Azure, and others, these providers' profit margins remain high.

On the purpose of decentralized cloud services: decentralized training vs. decentralized reasoning

coderworld:Machine learning is roughly divided into training (training) and reasoning (inference) Gensyn's P2P computing resources areIn which part does it come into play?

Ben:

Our focus isTraining, which involves the refinement of values.

trainFrom initial learning to fine-tuning,Inference involves only querying the model with data without changing it, basically seeing what the model predicts based on the input.

• Training requires a lot of computing resources, is usually asynchronous, and does not require immediate results.

• In contrast, inference needs to execute quickly to ensure user satisfaction in real-time applications, in stark contrast to the computationally intensive nature of training.

Decentralization is not yet sufficient to solve latency issues that are critical for inferenceIn the future, to perform inference efficiently, models need to be deployed as close to users as possible, minimizing latency by leveraging geographic proximity.

However, launching such a network is challenging because its value and effectiveness grow with the size of the network, which is consistent withMetcalfe's Law, similar to the development dynamics we have seen in projects like the Helium Network.

It is therefore unrealistic for Gensyn to address the inference challenge directly;This task is better suited to an independent entity focused on optimizing latency and network coverage.

We support a focus onProtocol optimized for a single function, rather than trying to develop in multiple areas simultaneously to avoid dilution of effectiveness. This specialization drives competition and innovation and leads to a series of interoperable protocols that are each proficient in a specific aspect of the ecosystem.

Ideally, in addition to running Gensyn nodes for computational tasks, users will be able to operate other functional nodes such as reasoning, data management, and data labeling. The interconnection of these networks will help build a powerful ecosystem in which machine learning tasks can be seamlessly transferred between various platforms. This decentralized vision of the future heralds a new layer of networks, each of which enhances the capabilities of machine learning through collective contributions.

Decentralized AI ecosystem: How to achieve a win-win situation with decentralized data protocols?

coderworld:Given that computation and data are important inputs to machine learning, how do Gensyn’s computation protocols work with data protocols?

Ben:

Computing is just one aspect; data is another important area where the same value flow model can be applied, although the verification and incentive mechanisms are different.

We envision a rich ecosystem of multiple nodes running on devices like your MacBook. You may have a Gensyn compute node, a data node, or even a data labeling node on your device, contributing to data labeling through gamified incentives or direct payments, often without direct awareness of the processes behind these models.

This ecosystem paves the way for what we ambitiously call the machine intelligence revolution, marking a new phase or evolution of the Internet. The current Internet serves as a vast repository of human knowledge in the form of text.

Computing is an important part, and data is another key area. The value flow model can also be applied, although the verification and incentive mechanisms are different.

We envision a vibrant ecosystem that includes multiple nodes running on devices like your MacBook. You might have more than one Gensyn compute node running on your device.May also include data nodes and data tag nodesThrough gamified incentives or direct payments, these nodes will contribute to data labeling, and users are usually unaware of the complex process behind it.

This ecosystem paves the way for what we call the machine intelligence revolution and marks a new stage in the development of the Internet, which is currently a vast textual repository of human knowledge.

The future Internet we envision is one that presents the Internet through machine learning models rather than text. This means that across the globe, from MacBooks to iPhones to cloud servers, pieces of machine learning models will be distributed across these devices, allowing us to query and reason through this distributed network. Compared to a centralized model controlled by a few cloud providers, this model promises a more open ecosystem, thanks to blockchain technology.

Blockchain not only facilitates resource sharing, but also ensures instant verification of tasks, verifying that tasks on remote devices are executed correctly and have not been tampered with.

Gensyn is committed to developing computational foundations within this framework and encouraging others to explore incentive schemes for data networks. Ideally, Gensyn will integrate seamlessly with these networks to improve the efficiency of machine learning training and applications.Our goal is not to monopolize the entire machine learning ecosystem, but to establish Gensyn as a protocol that optimizes the utilization of computing resources, sitting on top of electricity and significantly improving humanity's ability to use computing resources efficiently.

Gensyn specifically addresses the challenge of turning values and data into model parameters. Essentially, if you have a sample of data — whether it’s an image, book, text, audio, or video — and you want to turn that data into model parameters, Gensyn facilitates that process. This enables the model to make predictions or inferences about similar data in the future, evolving as the parameters are updated.The entire process of refining data into model parameters is exactly what Gensyn specializes in.,Other aspects of the machine learning stack are managed by other systems.

Additional topic: AI andcryptocurrencyAre startups limited by geography?

coderworld:Given your breadth of experience, could you compare how your early days as a builder and researcher in the tech field, dealing with the frustrations and challenges of computing and technology, compare to your current experiences? Can you share how this transition and London’s tech culture has impacted your development and achievements?

Ben:

The technology environment in London and the UK as a whole is significantly different from that in Silicon Valley.CommunityIt is full of brilliant people and groundbreaking work, but it tends to be more inward-looking. This creates barriers for newcomers trying to fit in.

I believe this difference stems from the contrasting attitudes between the UK and the US. Americans generally display a more open-minded outlook, while the British are generally more skeptical and conservative. This cultural nuance means that integrating and adapting to the UK tech ecosystem takes effort and time. However, once you do, you’ll find a vibrant and richCommunity, working on fascinating projects. The difference is in visibility and outreach; unlike Silicon Valley, where achievements are celebrated loudly, London’s innovators tend to work more quietly.

Lately, the UK, in particular, appears to be carving out a niche for itself in its shift toward decentralization and AI. This is partly because of regulatory developments in the US and Europe. For example, recent US regulations, as outlined in President Biden’s executive order, impose several restrictions on AI development, including mandatory government reporting for projects that exceed certain thresholds. These regulations could dampen enthusiasm among new developers. In contrast, the UK appears to be taking a more open approach, favoring open source over strict regulation, thereby fostering an environment more conducive to innovation.

San Francisco, known for its strong open source movement, faces new challenges with California legislation that echoes federal executive orders. These regulations, while intended to protect society, have inadvertently concentrated AI development in established entities. These entities have the ability to comply with regulatory requirements, while smaller players with potentially revolutionary ideas are disadvantaged. The United Kingdom recognizes the value of open source as a means of social oversight of AI development, avoiding the need for restrictive government surveillance. Open source practices naturally promote scrutiny and collaboration, ensuring that AI technology remains in check without stifling innovation.

The EU's initial regulation of AI was stricter than what we've seen in the UK, which has managed to find a balance to encourage open source development. This strategy not only aims to achieve the same regulatory goals, but also ensures that the market remains dynamic and competitive. The UK is particularly well positioned to foster a vibrant and open AI and crypto innovation ecosystem. This is an exciting time for London's tech industry.

Related reading:

1. https://docs.gensyn.ai/litepaper

2. https://epicenter.tv/episodes/471

3. https://www.techflowpost.com/article/detail_14995.html

4. https://hyperobjects.substack.com/p/understanding-maximum-computing-supply

The article comes from the Internet:Interview with Gensyn co-founder Ben Fielding: How can the decentralized computing protocol led by a16z achieve the democratization of AI?

Related recommendations: AI TokenThe other side: Most projects are busy with financial benefits rather than real-world impact

The naive idea of "token incentives + market direction" prevails here. Written by: Gagra Translated by: Xiaobai Navigation Coderworld This article summary This is not another VC article optimistic about the "AI + Web3" field. We are optimistic about the merger of these two technologies, but this article is a call. Otherwise, this optimism will eventually lose its reason. For…