OpenAI CEO Sam Altman在Dev Day活动周二公布了Decisions API，这是一款能够为AI模型提供预定义选项分类的工具1。该API在功能上与TypeSafe AI公司本月早些时候发布的Jev模型相似1，可用于图像分类和代理行为选择1。
Decisions API和Jev都被设计用于监控和保护AI代理的行为，以防止其在互联网上失控1。Altman表示："By focusing the model on that choice, we can make it extremely fast while keeping capabilities like image understanding, broad language support, and safety protections"1。使用Jev进行代理行为监控的成本为$2.94，相比frontier LLM的$372要低得多1。TypeSafe AI的CEO Diogo Almeida是前OpenAI工程师，曾共同发明了强化学习技术1。
OpenAI introduced its Decisions API during the company's Dev Day event, a tool designed to constrain and monitor the behavior of autonomous AI agents 1. The API functions similarly to Jev, a model recently released by TypeSafe AI, by offering predefined classification options that enable faster and more cost-effective decision-making for the Luna model in tasks such as image classification and agent behavior selection 1.
The new API aims to address concerns about runaway autonomous agents operating on the internet by providing safety guardrails through focused model constraints 1. CEO Sam Altman explained the approach during the announcement, stating: "By focusing the model on that choice, we can make it extremely fast while keeping capabilities like image understanding, broad language support, and safety protections" 1. The cost advantage is significant—using Jev-style monitoring for agent behavior control costs $2.94 compared to $372 for frontier large language models 1. TypeSafe AI's CEO Diogo Almeida, a former OpenAI engineer who co-invented reinforcement learning techniques, developed the competing Jev model that was released earlier this month 1.
评论
还没有评论，欢迎留下第一条。