Voices
Simon Willison's verdict on Jev: the output is a single float, and that makes hidden bias undetectable
Willison's September 21 post frames Jev as "unstructured state in, typed probabilistic decisions out" and calls it great for classification work like spam detection, labeling, prioritization and ranking at $0.042 per million input tokens with free output. His main objection is opacity: "put in all the text you want, the only thing you're going to get back is a floating point number," with no reasoning trace to audit. He demonstrates the risk by asking Jev to rate Bay Area cities on a "good city" metric, getting Cupertino at the top and East Palo Alto at the bottom, and warns explicitly against using it for employment screening.
↳ Follow the thread