Many AI tasks require choosing among predefined options rather than generating an open-ended response. Decision models score these options, providing an explicit signal of uncertainty to guide automation and human review. When validated, these scores help reduce the risk of acting on unsupported conclusions. Compact models can run locally on a CPU, reducing hardware requirements, power consumption, and operating costs.
This session explores their architectures, main types, and differences from generative LLMs, with practical examples covering document classification, sensitive information screening, request routing, and model or tool selection.