Darren LI

整个sim-to-real过程如图4所示,共分为四步:

(1)识别出机器人的物理参数,并对机器人进行刚体运动学/动力学建模;

(2)收集真实的关节电机执行数据,训练一个Actuator Net;

(3)在仿真中,利用Actuator Net建模关节电机,并结合第一步中的刚体运动学/动力学建模,进行强化学习;

(4)将第3步中训练得到的策略部署到真机上。

But regular end-users, even engineers, cannot assess safety based on model weights. And prompt injection remains a large attack vector.

This has some problems: - Failure rate is high - Even when successful, they're often costly + inefficient - Full autonomy removes humans from being in control (!) - Agentic controls are weak - Security concerns are very real. Autonomy + access to your private data =

具体到Bolt,因为单笔金额灵活、数量会比较多,我告诉LP,可以把Bolt看成一个一揽子计划(basket),最终的总体投资额(包括后续加码)会在2000万到5000万美元。这些项目中最终能几个有在contextual data、personal data、collaborative data这三个方面实现积累并且与时俱进的团队,最终的收益也不会差。而且, 未来应用公司创造的总价值会远超过大模型。