Startup TypeSafe AI has launched Jev, an AI model designed to provide narrow judgments and probabilities rather than generating text. Co-founder and CEO Diogo Almeida, a former OpenAI researcher and co-author of the InstructGPT paper, leads the company. The model operates by scoring predefined options provided by developers.
Jev functions by analyzing input text and assigning labels based on programmed criteria. For example, in customer service, it can categorize messages as payment issues or refund requests. The software then applies fixed rules to route the request accordingly. This process allows for automated sorting without the need for traditional chatbot responses.
The model achieves response times between 70 and 500 milliseconds by computing outputs in parallel and skipping text generation. TypeSafe claims this speed is significantly faster than standard language models. The company suggests this performance enables real-time checks for AI assistants, such as verifying if a draft contradicts previous conversation history.
TypeSafe AI prices the model at $0.042 per million input tokens, with no charge for outputs. While the company states the model cannot hallucinate because it is restricted to preset options, it notes that factually incorrect choices within those options remain possible. Developers can currently access the model via a waitlist.
