Diogo Almeida, once a pivotal figure at OpenAI and co-developer of initiatives like GPT-4, has recently unveiled a novel AI model category known as Jev. Distinct from conventional AI models that primarily generate text or facilitate chats, Jev is adept at providing typed responses and precise probability values tailored to the input data and structured queries. In classification tasks, Jev's prowess notably outshines that of the GPT series models. Within less than a month of its launch, the model has already been appraised at a staggering $10 billion and is presently engaged in discussions to secure $1 billion in funding. Diogo Almeida contends that prevailing products such as ChatGPT and Claude Code, which revolve around Reinforcement Learning from Human Feedback (RLHF), are merely supplementary AI models crafted to cater to human inclinations and fall short of attaining genuine automation. He underscores that the RLHF phase signifies a diversion for the industry and that the initial Scaling Law is not entirely precise. The forthcoming era of AI will prioritize automation and will possess the capability to generate more intelligent software. To realize this vision, his company, TypeSafe, is delving into a novel post-training approach to refine calibrated decision-making abilities and to construct an AI technology framework explicitly tailored for automation.