A New AI Model from Ex-OpenAI Researcher Aims for Efficiency and Automation
Diogo Almeida, former OpenAI researcher who helped develop ChatGPT's reinforcement learning technique RLHF, has launched a new model called Jev through his startup TypeSafe AI. Unlike traditional large language models (LLMs), Jev does not generate text but instead produces calibrated probabilities for decision-making tasks. This approach makes the model faster and cheaper to use while avoiding hallucinations due to predefined outputs. Developers are showing strong interest in Jev, with some reporting significant improvements in speed and accuracy compared to existing LLMs like OpenAI's Luna 5.6 or Google’s Gemini when used for classification tasks such as email sorting or command safety checks. The model is also being explored for monitoring other AI agents at lower cost. Jev aims to enable widespread, distributed use of intelligent software by reducing the costs associated with deploying AI in automation workflows and real-time decision-making systems.
Topics: AI automation LLM alternatives reinforcement learning