TypeSafe's Jev is a genuine breakthrough – snap judgments with calibrated probabilities instead of generated text. My bet is OpenAI is already figuring out how to copy it, and then embed it inside its own models where Jev can't follow.