Andrew Trask describes combining model outputs and argues that ensembles and routingAI model routing sends each request to a model chosen for that request's complexity, cost, speed, privacy, or other requirements. can improve performance or cost in some settings. He connects the idea to earlier machine-learning practice and to a possible market for specialized capabilities.
Andrew Trask discusses privacy-preserving computationData privacy governs how personal, confidential, or sensitive information is collected, used, shared, retained, and protected in AI systems. and proposes a way for data owners to contribute to predictions without distributing raw data. He distinguishes local client models from knowledge held by servers in his proposed architecture.
The conversation covers benchmark saturationEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria., trust and incentives. His predictions about technical deployment and future markets are presented as proposals, not guarantees.
Watch on YouTube
