Eye on AI: Alexander Whedon of Subquadratic on Solving AI's Quadratic Compute Problem
Alexander Whedon, co-founder and CTO of Subquadratic, explains how their sparse attention mechanism solves the quadratic compute problem in AI models, where doubling the context window quadruples compute. SubQ's method achieves 40 times faster inference and 64 times less compute than standard attention at one million tokens, without quality tradeoffs. This innovation allows enterprises to process hundreds of billions of tokens of data that were previously inaccessible due to cost and complexity. Whedon also discusses how this technology transforms coding agents, enabling them to read entire codebases efficiently and perform a fundamentally different, longer horizon of work. The conversation covers the impact on long context reasoning, the potential end of RAG as a complex workaround, and why SubQ is uniquely positioned in multi-million token pre-training.
Original sources
YouTube Eye on AI
Video:86% of What Coding Agents Do Is Just Reading — Not Solving | Alexander Whedon of Subquadratic ↗
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.