TypeSafe AI launches Jev decision model for automation

TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, has launched Jev, a new AI model categorized as a ‘System One’ model. Unlike traditional Large Language Models (LLMs) that generate conversational text, Jev is designed specifically for high-speed, low-cost automated decision-making. It processes unstructured input and returns structured, typed outputs such as choices, scores, or boolean values accompanied by confidence probabilities.

Jev aims to solve the latency and cost issues associated with using general-purpose LLMs for routine tasks like data classification, risk assessment, and workflow routing. By eliminating the text-generation step, the model can respond in 70 to 500 milliseconds. The company reports that Jev is significantly more cost-effective than standard LLMs, with input costs cited at approximately $0.042 per million tokens and output tokens being free.

Early adoption has been noted in various sectors, including software development and cryptocurrency trading. For instance, Vercel reported that replacing an OpenAI model with Jev for security classification increased speed by 5 to 18 times. The model uses a training method called ‘Reinforcement Learning for Calibrated Decisions’ (RLCD) to ensure outputs remain within defined parameters, effectively preventing the ‘hallucinations’ common in generative AI.