Intrigued by the claims of emergent reasoning capabilities in LLMs trained on general web corpora, in this paper, we set out to investigate their planning capabilities.

Security experts have urged people to be cautious with the viral agentic AI tool, known for being highly capable but also wildly unpredictable. Last month, Jason Grad issued a late-night warning to the 20 employees at his tech startup. "You've likely seen Clawdbot trending on X/LinkedIn. While cool, it is currently unvetted and high-risk for our environment, he wrote in a Slack message with a red siren emoji. "Please keep Clawdbot off all company hardware and away from work-linked accounts." Grad isn't the only tech executive who has raised concerns to staff about the experimental agentic AI tool, which was briefly known as MoltBot and is now named OpenClaw. A Meta executive says he recently told his team to keep OpenClaw off their regular work laptops or risk losing their jobs. The executive told reporters he believes the software is unpredictable and could lead to a privacy breach if used in otherwise secure environments. He spoke on the condition of anonymity to speak frankly.

In this supplementary material, we first give additional details about the method in Sec. 1. Queries used for zero-shot semantic segmentation. We do this for all the annotated classes in the dataset (second column). One can see that, for example, class name'manmade' lacks descriptive specificity. In the text description of this class, we can find "... buildings, walls, guard rails, fences, poles, street signs, traffic lights ..." and more. Table 1: Queries used for zero-shot semantic segmentation.