Jev, a non-language AI model from a ChatGPT co-inventor, is winning over developers

Diogo Almeida, a former OpenAI researcher who helped build ChatGPT and invented reinforcement learning from human feedback, has left the company to found TypeSafe AI, which just released Jev — a transformer-based model that is not a large language model. Instead of text, Jev outputs probabilities the company calls "calibrated decisions," with user-defined outputs that make hallucination impossible. Because it eschews language, the model is extremely cheap and fast: output tokens are free and input tokens are metered by the billion rather than the million, offering developers a cheaper path to software automation.