The race to scale GPU compute and the A.I. Renaissance with Gensyn Co-Founders | EP #106
Oct 19, 2023
auto_awesome
Ben Fielding and Harry Grieve, Co-Founders of Gensyn, discuss the race to scale GPU compute and the AI renaissance. They explore Gensyn's vision of a global supercluster accessible to anyone, the limitations of decentralized compute, the cost comparison between decentralized and traditional compute clusters, and the importance of owning your own AI. They also touch on the verification problem in decentralized systems and the impact of decentralization on the AI ecosystem.
Gensyn aims to automate access to computational resources for machine learning training, democratizing access and optimizing usage of existing chips.
Gensyn envisions becoming the network for machine learning compute, serving as the infrastructure for the upcoming AI-driven industrial revolution.
Decentralization in Gensyn is enabled through crypto technologies, solving the verification problem and allowing decentralized interactions without relying on trusted centralized entities.
Deep dives
Jensen's Origin: Solving the Compute Problem in Machine Learning
Jensen was founded with a focus on solving the compute problem in machine learning. The co-founders, Ben and Harry, realized that the design of neural networks was being done manually, and machine learning experiments were limited by the availability of computational resources. They recognized that machine learning was computationally expensive and inaccessible to many researchers and developers. Their vision was to automate the process of accessing compute for machine learning training, optimizing usage of existing chips and democratizing access to compute for the entire machine learning community.
The Vision of Jensen: Building a Network for Machine Learning Compute
Jensen's vision is to be the network for machine learning compute, serving as the substrate for the next industrial revolution driven by AI. By pooling together unused and underutilized compute resources, Jensen aims to provide affordable and scalable access to computational power for training machine learning models. The network is designed with an open and agnostic approach, allowing various use cases and specialized models to be built on top. Jensen's goal is to be the underlying infrastructure that powers machine learning advancements, increasing impact and enabling the development of autonomous systems.
Training-Focused and the Importance of Verification in Jensen's Network
Jensen focuses primarily on machine learning training compute, as training asynchronous processes fit well with decentralization. The network aims to optimize usage of existing compute resources by automating the process and eliminating the need for centralized, costly cloud providers. Decentralization in Jensen is enabled through crypto technologies, by solving the verification problem and allowing peer-to-peer interactions without the need for trust in centralized entities. While some might argue about the cost of bandwidth, Jensen takes a long-term perspective, emphasizing the historical advancements of machine learning and countering concerns about bandwidth costs. Jensen envisions a future where specialized hardware and a global network of devices support efficient and cost-effective training for a wide range of machine learning models.
Decentralization and cost reduction in compute: a game changer
The podcast episode explores how decentralization and cost reduction in compute can revolutionize the industry. The guest speaker highlights that by making compute resources more accessible and affordable, organizations can level the playing field and compete with larger players. The focus is on reducing the base cost of compute to the fair market value, making it equivalent to open-source access. This value proposition appeals not only to those interested in crypto but also to machine learning engineers in the web2 world. The speaker emphasizes the importance of unlocking these value propositions through decentralization, enabling significant cost savings and attracting a wider user base.
Addressing challenges in decentralized GPU compute
The podcast episode delves into the challenges of decentralized GPU compute and shares insights on addressing those obstacles. One major challenge is the verification problem, which involves confirming the validity of machine learning models. To address this, the guest speakers discuss a probabilistic, game-theoretic approach combined with secure cryptographic techniques. They mention the importance of reproducibility in training machine learning models and ways to overcome determinism issues related to different devices. Additionally, they highlight the need for an efficient verification process, considering bandwidth constraints and the granularity of work sections. The goal is to achieve a secure, practical solution while remaining adaptable as technology and cryptographic techniques evolve.
Ben Fielding and Harry Grieve are the Co-Founders of Gensyn - The Gensyn network is the Machine Learning Compute Protocol that unites all of the world’s compute into a global supercluster, accessible by anyone at any time
Ben Fielding Twitter: https://twitter.com/fenbielding
Harry Grieve Twitter: https://twitter.com/_grieve
Gensyn's Website: https://twitter.com/gensynai
Logan Jastremski's Twitter: @LoganJastremski
Frictionless's Twitter: @_Frictionless_
Frictionless's Website: https://frictionless.fund/
___
Timecodes:
0:00 - Intro
1:10 - Founders Background
10:20 - Gensyn's Vision
15:20 - AI is the next industrial revolution
21:20 - Machine Learning and Training Compute
32:00 - Real World Limitations of Decentralized Compute
39:30 - Where does Compute come from?
50:50 - Decentralized Cost vs. Traditional Compute Clusters
55:15 - Owning your own AI
1:01:00 - Verification problem
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
