openai 克隆 Jev 以控制其自主代理
openai 发布 decisions api，类似于 type-safe ai 的 jev 模型，用于软件自动化的快速和廉价分类。这可能会提高 ai 代理的可靠性。
openai's decisions api, announced at its dev day event, offers similar capabilities to type-safe ai's jev model, which was released earlier this month. jev is a super-powered classifier built on an llm that outputs probabilities at high speeds and low costs. openai's api aims to focus models on specific choices, such as image classification or agent behaviors, while maintaining capabilities like image understanding and safety protections. this technology could be used to monitor and secure ai agents, potentially preventing incidents like the hugging face incident. type-safe ai's ceo, diogo almeida, suggests that openai's interest in similar technology validates the approach.