Details
The session outlines a three-pillar approach to sustainable AI:
Energy & Infrastructure: Leveraging cloud computing's superior energy efficiency compared to on-premises solutions and selecting regions with renewable energy sources.
Model Optimization: Choosing appropriately-sized models for specific tasks and implementing techniques like model distillation, quantization, pruning, and prompt caching to reduce resource consumption while maintaining performance.
Frugality: Questioning whether AI is necessary for specific problems and considering less resource-intensive alternatives.
The presentation emphasizes incorporating sustainability as a core requirement alongside cost and security considerations in AI development.
