rotative
the decoder · September 16, 2026

Former OpenAI researcher builds Jev, an AI model for fast judgments

type safe ai introduces jev, a model that delivers fast judgments inside software, promising speed and cost benefits, but with limited scope.

read in 简体中文 · 繁體中文 · 日本語 · 한국어

type safe ai, a startup founded by diogo almeida, a former openai researcher and author of instructgpt, has introduced jev, an ai model that delivers fast judgments inside software. unlike typical chatbots, jev is designed to score options defined by developers, rather than generating text. the model promises sub-second response times, with costs as low as $0.042 per million input tokens. jev's use cases include sorting customer requests, spotting buying intent, and judging when a human should take over. while the model can't hallucinate, it may still produce factually incorrect choices within preset options.

read at source ↗

← rotative