Raymond Weitekamp describes recursive decision models that divide context outside the model and allow a decision to call another decisionA decision model selects or scores possible answers for a defined decision.. He proposes that such systems should work without generative models while also being able to compose with them.
He demonstrates hierarchical search, visual localization and a Factorio example, noting that some are simplified demonstrations. He explains a shared map and an arbiter for coordinating model decisions, then shows an experimental voice interface that learns when to trust a smaller, faster modelAI model routing sends each request to a model chosen for that request's complexity, cost, speed, privacy, or other requirements..
Watch on YouTube
