Zero-shot LLM agents diverge from human experts in dialog acts, tool usage, and knowledge synthesis; this behavior gap widens with task complexity and correlates with lower task performance.
The agent provides information about the hotel, such as its name, location, price, rating, etc
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
The Behavior Gap: Evaluating Zero-shot LLM Agents in Complex Task-Oriented Dialogs
Zero-shot LLM agents diverge from human experts in dialog acts, tool usage, and knowledge synthesis; this behavior gap widens with task complexity and correlates with lower task performance.