Scout AI is using technology borrowed from the AI industry to power lethal weapons--and recently demonstrated its explosive potential. Like many Silicon Valley companies today, Scout AI is training large AI models and agents to automate chores. The big difference is that instead of writing code, answering emails, or buying stuff online, Scout AI's agents are designed to seek and destroy things in the physical world with exploding drones. In a recent demonstration, held at an undisclosed military base in central California, Scout AI's technology was put in charge of a self-driving off-road vehicle and a pair of lethal drones. The agents used these systems to find a truck hiding in the area, and then blew it to bits using an explosive charge.

In supervised learning, the question of data quality and curation has been overshadowed in recent years by increasingly more powerful and expressive models that can ingest internet-scale data.

Learning from active human involvement enables the human subject to actively intervene and demonstrate to the AI agent during training. The interaction and corrective feedback from human brings safety and AI alignment to the learning process.

In this paper, we present a hyperparameter-free offline safe IL algorithm, SafeDICE, that learns safe policy by leveraging the non-preferred demonstrations in the space of stationary distributions. Our algorithm directly estimates the stationary distribution corrections of the policy that imitate the demonstrations excluding the non-preferred behavior.

Large language models have shown tremendous performance in a variety of tasks.

Recent work has shown that language models' (LMs) prompt-based learning capabilities make them well suited for automating data labeling in domains where manual annotation is expensive.

This procedure can leverage noisy, asynchronous human feedback to learn policies with no hand-crafted reward design or exploration bonuses.