Former OpenAI Researcher's Jev AI Model Promises Dramatic Speed and Cost Cuts

TypeSafe AI emerged from stealth this week with Jev, a radically different approach to AI decision-making that claims to run 193 times faster and cost 445 times less than frontier language models on comparable tasks. Built by Diogo Almeida, a former OpenAI researcher who spent four years developing reinforcement learning from human feedback for ChatGPT and GPT-4, Jev abandons the token-by-token text generation that defines modern LLMs in favor of a "System One" architecture inspired by cognitive science.