Diogo almeida's jev model offers structured decision-making without text generation

TypeSafe AI, founded by Diogo Almeida, a former OpenAI researcher involved in RLHF development, has launched Jev, a new AI model designed for structured decision-making. Jev, dubbed the 'System One Model,' does not generate text but instead provides direct, structured outputs. The model is optimized for fast, low-latency decision-making in repetitive tasks, such as scoring or binary classification. It uses a technique called 'calibrated decision reinforcement learning' to improve the accuracy of its confidence estimates. Jev is claimed to be 20 to 200 times faster and 40 to 400 times cheaper than leading large language models for structured tasks. Input tokens cost $0.042 per million, while output tokens are free. The company raised $40 million in seed funding led by DCVC. Jev's performance was demonstrated in a Doom gameplay video and a Wikipedia navigation benchmark, outperforming other high-speed models.