Dan Shipper

OpenAI's new model Strawberry represents a major leap in reasoning abilities for language models. Unlike current models that rely on statistical patterns, Strawberry is trained through 'process supervision' - rewarding correct step-by-step reasoning. This allows it to solve novel math problems and puzzles it hasn't encountered before, hallucinating less. Originally built to generate training data for OpenAI's upcoming model Orion, a smaller version may be integrated into ChatGPT, significantly upgrading its abilities. Strawberry's existence implies future AI models could achieve human-level reasoning through data focused on logic and chain of thought processes.