Jon Saad-Falcon

Local inference using small language models (≤20B parameters) can accurately answer 88.7% of real-world queries while achieving 5.3x improvement in intelligence-per-watt efficiency from 2023-2025, enabling meaningful redistribution of demand from centralized cloud infrastructure.

“Author of "Intelligence Per Watt: Measuring Intelligence Efficiency of Local AI" on arXiv”

“Co-author of article discussing frontier model capabilities and appropriate use cases for high-intelligence AI systems.”