A New Kind of Model for AI Decision-Making?

A new model called Jev from TypeSafe.AI is introduced as a 'System One' model designed for AI decision-making rather than text generation. Unlike LLMs trained with RLHF, Jev uses Reinforcement Learning for Calibrated Decisions (RLCD) to return structured answers with associated confidence probabilities. A practical comparison was conducted against OpenAI's models for a banking intent classification task. While Jev demonstrated significantly faster performance, it achieved lower accuracy than the OpenAI models, though its confidence scores proved to be well-calibrated.