TypeSafe AI launches Jev, a fast model for software automation
TypeSafe AI has released Jev, a transformer-based model designed for software automation rather than text generation. The company was founded by Diogo Almeida, a former OpenAI researcher who helped develop ChatGPT and reinforcement learning from human feedback (RLHF). Jev returns probabilities, which TypeSafe calls “calibrated decisions,” instead of language. Because users define the possible outputs in advance, the company says the model cannot hallucinate. Its output tokens are free, while input is metered by the billion rather than the million.
Early users are testing Jev for classification and other automated tasks. Pranit Sharma of Vercel said replacing OpenAI’s ChatGPT Luna 5.6 with Jev made command-safety checks 5 to 18 times faster and more accurate. Nikhil Mudholkar of Bryo AI found Gemini slightly more accurate for business-email classification, but 10 to 20 times more expensive; he highlighted Jev’s probability scores.
The model may also monitor LLM agents, detect jailbreaks, or route requests to suitable models. Its architecture has not been disclosed, although outside observers suspect it may use an open-weight LLM. Almeida says Jev is trained only on synthetic data through “reinforcement learning from calibrated decisions.” TypeSafe plans additional versions for new modalities.
Written by Sakhanda Wire AI from reporting by TechCrunch. Facts, quotes and full context are in the original article below.