A new AI model from a ChatGPT pioneer offers probabilistic insights for automation

Almeida, a former OpenAI researcher who contributed to the development of reinforcement learning from human feedback (RLHF), has launched a new AI model called Jev. This transformer-based model, developed by his startup TypeSafe AI, does not generate text but instead provides calibrated probabilities, making it ideal for automation. Jev is designed to be faster and cheaper than traditional large language models (LLMs), with input tokens metered by the billion rather than the million. Developers have shown strong interest, with the company temporarily unable to serve users due to high demand. Jev’s ability to provide confidence scores and avoid hallucination has made it appealing for tasks like email classification and workflow automation. Almeida, who left OpenAI two years ago, believes Jev could revolutionize how intelligence is deployed, drawing inspiration from the Jevons paradox. The model is trained exclusively on synthetic data using a technique he calls reinforcement learning from calibrated decisions. Almeida envisions a future where smart software becomes ubiquitous, much like the early internet, rather than relying on mega apps.