Can LLMs Separate Pasted Artifacts from User Speech? Absorption at Unmarked Prompt Seams

Large language models (LLMs) receive each user message as plain text, even when it combines text from different sources. For example, a user may paste text into a prompt and keep typing a comment directly below it.