Nvidia的研究人员开发了名为Eureka的AI代理，它可以使用生成式AI和自主奖励算法来教授机器人复杂的技能。Eureka的方法比人类编写的程序高效50%以上，使机器人能够学会像转笔花招之类的任务。这一技术已经被用于教授各种机器人近30项不同的任务。Nvidia还推出了SteerLM，通过对话和用户反馈来培训AI助手以改善对齐。Eureka和SteerLM都巧妙地利用了先进的神经网络。这些创新标志着机器人灵巧和AI互动方面的重大进展。
#NVIDIA #AI #Eureka

#Nvidia researchers have developed #Eureka , an #AI agent that can teach robots complex skills using generative AI and autonomous reward algorithms. Eureka's approach is over 50% more effective than human-authored programs, enabling robots to learn tasks like pen-spinning tricks. The technique has been used to teach various robots nearly 30 different tasks. Nvidia also introduced SteerLM, which trains AI assistants to improve alignment through conversations and user feedback. Both Eureka and SteerLM utilize advanced neural networks creatively. These innovations mark significant progress in robotic dexterity and AI interaction.