Saved report copy
Recorded
ChatGPT inventor's new non-language model excites developers
TechCrunch reports that former OpenAI researcher Diogo Almeida left to start TypeSafe AI and released Jev, a transformer based model that outputs calibrated decisions instead of text. Developers are responding because Jev is cheap and fast, cannot hallucinate when outputs are predefined, and returns confidence scores useful for automation. In tests cited by the outlet, Vercel saw results five to 18 times more quickly with greater accuracy after replacing an OpenAI model, while another test found Jev far cheaper than Gemini for email classification.