Intent Mismatch Causes LLMs to Get Lost in Multi-Turn Conversation

Geng Liu, Fei Zhu, Rong Feng, Changyi Ma, Shiqi Wang, Gaofeng Meng · 2026

Read Intent Mismatch Causes LLMs to Get Lost in Multi-Turn Conversation by Geng Liu, Fei Zhu, Rong Feng, Changyi Ma, Shiqi Wang, Gaofeng Meng (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07338.txt.

Description

Language models famously get worse in multi-turn conversation than with a fully specified single prompt, a drop called Lost in Conversation. Geng Liu and five co-authors argue the cause is not model weakness but an intent match gap between user and model.

They show the drop is a breakdown between user and model, not capability. Argue theoretically that bigger models or better training alone cannot close it, because the gap comes from structural ambiguity in conversational context. The paper reframes a familiar frustration: follow-up questions refine intent gradually, and systems must track that refinement. Dialogue teams get a diagnosis that points at chat design, not just scale.

Catalog record

CREATORGeng Liu, Fei Zhu, Rong Feng, Changyi Ma, Shiqi Wang, Gaofeng Meng
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07338

Browse the full library · About the Tower of Babel