Decentralizing Compute for Machine Learning
summary
Gensyn builds a permissionless peer-to-peer ML compute protocol that claims lower unit costs than AWS by pooling idle GPUs into an internet-scale supercluster, already running an internal network used to train small models; core work targets verifiable training with only double-digit overhead, reproducible results across different GPU hardware, optimal model sharding, and a data availability layer (current blocker: storage deals like 180-day Filecoin minimum). Roadmap: private testnet this year, public testing next year, plus ongoing security and distributed-training research.
AI-written summary · about
chapters (38)
- 0:00Introduction to Gensyn and Decentralized ML Compute
- 0:50The Growing Intersection of Crypto and AI
- 1:20Gensyn's Mission: Internet-Scale ML Training
- 1:57Founding Story: The GPU Access Problem
- 2:29Rising Compute Costs and GPT-2's Impact
- 3:02Academic vs Industry ML Resource Disparity
- 3:30Vision for Internet-Scale Machine Learning
- 4:01ML as the Next Industrial Revolution
- 4:31Transition to Dynamic AI-Powered Internet
- 5:02Controlling the Future of Online Information
- 5:32Training Models Across Any Compatible Device
- 6:03Five Key Research Challenges in Decentralized ML
- 6:33Reproducibility Across Different Hardware
- 6:58Optimal Model Sharding and Distribution
- 7:26Blockchain and Data Availability Considerations
- 7:51Current Status: Internal Network Testing
- 8:27Implementation Progress and Security Focus
- 8:55Data Availability Layer Requirements
- 9:52Four Key DA Needs for ML Protocols
- 10:22Temporary vs Permanent Storage Solutions
- 11:15Decentralized Model Training Research
- 11:45Physical Limits of Centralized Compute
- 12:18Breaking the Data Center Scaling Barrier
- 12:46Three Core Focus Areas for Gensyn
- 13:16Permissionless Network Principles
- 13:49Machine-to-Machine Compute Demand
- 14:23Autonomous AI Economic Agents
- 14:56The Liberty Argument for Decentralized AI
- 15:26Owning Your Artificial Neurons
- 15:51Crypto's Role in AI's Future
- 16:24Three-Layer Model for Crypto+AI Integration
- 16:50ML Ops: Attribution and Guardrails
- 17:23AGI on Cryptographic Substrates
- 17:49Four-Year Prediction for Permissionless AGI
- 18:23AI-Physical World Interface Challenges
- 18:55Gensyn's Roadmap: Private to Public Testing
- 19:27Research Opportunities in Distributed ML
- 19:56Need for External Data Solutions
related talks
09mDemocratizing Access to GPUs for Faster MLAI Innovation with Generative AI
13mDemocratizing AI with Open Source Compute
36mAccelerating High-Performance Machine Learning at Scale i...
31mKubernetes For GPU Powered Machine Learning Workloads In Academia
28mDePIN & Privacy
10mGenesis Startup Competition Pitch: NeuroMesh
14mPowering the Future of Decentralized AI
24mA Better and More Efficient ML Experience for CERN Users