Former OpenAI researcher builds an AI model that judges options instead of writing text
decoder ·
TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, has launched a new model named Jev that focuses on making rapid judgments rather than generating text. Unlike standard chatbots, Jev evaluates predefined options and assigns probabilities to help software systems sort tasks efficiently. For example, it can categorize customer service inquiries as payment issues or refund requests within milliseconds. The company claims this parallel processing approach allows for sub-second response times, significantly outperforming traditional language models in speed. While the model is marketed as being unable to hallucinate outside its preset categories, it can still make factual errors within those limited choices. TypeSafe AI highlights cost efficiency at $0.042 per million input tokens, though independent performance benchmarks are currently lacking. Developers must verify the reliability of these background judgments for their specific use cases before full adoption.
decoder