Skip to content
View original post on X: Artificial AnalysisOfficial
Published
31/100

Claude Opus 5.5 keeps adjusting a crocodile after declaring the task done

AIArtificial Analysis reports that Claude Opus 5.5, controlling a robot arm, declared a task complete after 23 minutes, saying "The crocodile looks fully inside now." It then spent 22 more minutes pushing the crocodile back and forth and ended almost where it started.
Artificial AnalysisVerified on XOfficial
@ArtificialAnlys
View on X

Knowing when to stop is hard for language models controlling a robot arm

➤ Claude Opus 5.5 completed the task in 23 minutes, and said so: "The crocodile looks fully inside now"

➤ It then spent 22 minutes pushing the crocodile back and forth, only to finish almost exactly where it was before

Source details

Artificial Analysis · x.comPublished · added here

Original title: Knowing when to stop is hard for language models controlling a robot arm

Topics

Anthropic / ClaudeEmbodied AIEvals & benchmarks