The first (public) System One Model; Jev gives AI the properties of code (typesafe.ai)

TypeSafe AI has introduced a new class of machine-native models, called System One Models, designed to overcome the limitations of traditional Large Language Models (LLMs) that rely on Reinforcement Learning from Human Feedback (RLHF). Unlike LLMs, which are tailored for human communication and often suffer from issues like mode dropping and overconfidence, System One Models utilize a new architecture and a novel training algorithm known as Reinforcement Learning for Calibrated Decisions (RLCD). This evolution allows these models to produce typed outputs with calibrated confidence levels, enabling better automation and reliability in decision-making. The significance of this advancement lies in its capacity to deliver rapid, cost-effective, and type-safe decisions for automated systems, with performance reportedly 193.6 times faster and 444.6 times cheaper than existing LLMs. The ability to define when a model should act autonomously versus when it should seek human review transforms how AI can be integrated into software workflows. With a pricing of $42 per billion tokens, TypeSafe AI's offering presents a promising alternative for developers seeking intelligent solutions that prioritize reliability and accuracy, addressing a critical gap in the current AI landscape dominated by human-interactive models.