TypeSafe AI Launches Jev, a Fast Non-Text-Generating System One AI Model
TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, emerged from stealth with $40 million in seed funding and launched Jev, a 'System One' AI model designed to deliver fast, structured decisions without generating natural language text. Jev introduces a paradigm shift away from traditional autoregressive LLMs for structured automation, promising 20 to 200 times faster execution and 40 to 400 times lower operating costs. This approach could drastically reduce infrastructure overhead for software workflows that require repetitive, high-frequency decision-making rather than conversational dialogue. Jev outputs three fixed structured primitives—Choice (selecting from up to 255 predefined options), Score (numerical ratings), and Noul (boolean probabilities)—and is trained using Reinforcement Learning for Calibrated Decisions (RLCD). Because Jev completely eliminates autoregressive generation, output tokens are effectively free, with input tokens priced at $0.042 per million.
## BACKGROUND
The concept of 'System 1' thinking stems from cognitive psychology, representing fast, intuitive, and automatic judgment, in contrast to the slow, deliberative reasoning of 'System 2'. Traditional Large Language Models (LLMs) operate by autoregressively predicting token after token to produce human-readable text, which incurs significant computational latency and expense for simple decision-making tasks. Founder Diogo Almeida previously contributed to Reinforcement Learning from Human Feedback (RLHF) at OpenAI, the key technique that enabled ChatGPT.