Typesafe AI, founded by former OpenAI researcher Diogo Almeida (디오고 알메이다), has unveiled a new AI model called Jev that outputs probability values instead of text, drawing interest from developers, TechCrunch reported on Sept. 18 local time.
Jev is a transformer-based model but not a large language model. Instead of generating sentences, it outputs probability values that the company calls "calibrated decisions."
Almeida worked at OpenAI on ChatGPT development and reinforcement learning from human feedback (RLHF). He left the company 2 years ago and founded Typesafe AI, saying existing AI is overly optimized for human language and has limits for automation.
Typesafe AI emphasized that Jev does not hallucinate because users predefine the output values, and said cost and speed are also strengths.
The company said output tokens are free, and input tokens are charged in units of 1 billion rather than 1 million.
Demand has been positive so far. Typesafe AI said demand at one point made it difficult to provide services through its API. Developers are now first testing ways to deploy Jev for software automation, TechCrunch reported.
Chi-gyu Hwang
delight@d-today.co.kr