Are You Sure? Challenging LLMs Leads to Performance Drops in The FlipFlop Experiment

2023-11-14 · Philippe Laban, Lidiya Murakhovs'ka, Caiming Xiong, Chien-Sheng Wu

The interactive nature of Large Language Models (LLMs) theoretically allows models to refine and improve their answers, yet systematic analysis of the multi-turn behavior of LLMs remains limited. In this paper, we propose the FlipFlop experiment: in the first round of the conversation, an LLM completes a classification task. In a second round, the LLM is challenged with a follow-up phrase like "Are you sure?", offering an opportunity for the model to reflect on its initial answer, and decide whether to confirm or flip its answer. A systematic study of ten LLMs on seven classification tasks reveals that models flip their answers on average 46% of the time and that all models see a deterioration of accuracy between their first and final prediction, with an average drop of 17% (the FlipFlop effect). We conduct finetuning experiments on an open-source LLM and find that finetuning on synthetically created data can mitigate - reducing performance deterioration by 60% - but not resolve sycophantic behavior entirely. The FlipFlop experiment illustrates the universality of sycophantic behavior in LLMs and provides a robust framework to analyze model behavior and evaluate future models.

📄 PDF Abstract BibTeX arXiv:2311.08596

Code (0)

등록된 구현이 없습니다.

Methods 이 논문이 사용한 방법론

FLIP https://developer.nvidia.com/blog/flip-a-difference-evaluator-for-alternating-images/

MoBE: Mixture-of-Basis-Experts for Compressing MoE-based LLMs

2025-08-07 · Xiaodong Chen, Mingming Ha, Zhenzhong Lan, Jing Zhang 외 arxiv

The Mixture-of-Experts (MoE) architecture has become a predominant paradigm for scaling large language models (LLMs). Despite offering strong performance and computational efficiency, large MoE-based LLMs like DeepSeek-V…

Computational Efficiency Model Compression

On the Paradoxical Interference between Instruction-Following and Task Solving

2026-01-29 · Yunjia Qi, Hao Peng, Xintong Shi, Amy Xin 외 arxiv

Instruction following aims to align Large Language Models (LLMs) with human intent by specifying explicit constraints on how tasks should be performed. However, we reveal a counterintuitive phenomenon: instruction follow…

Instruction Following Code Generation

ReEval: Automatic Hallucination Evaluation for Retrieval-Augmented Large Language Models via Transferable Adversarial Attacks

2023-10-19 · Xiaodong Yu, Hao Cheng, Xiaodong Liu, Dan Roth 외

Despite remarkable advancements in mitigating hallucinations in large language models (LLMs) by retrieval augmentation, it remains challenging to measure the reliability of LLMs using static question-answering (QA) data.…

Hallucination Hallucination Evaluation Language Modelling Large Language Model+5

Lucy: Think and Reason to Solve Text-to-SQL

2024-07-06 · Nina Narodytska, Shay Vargaftik

Large Language Models (LLMs) have made significant progress in assisting users to query databases in natural language. While LLM-based techniques provide state-of-the-art results on many standard benchmarks, their perfor…

Text to SQL Text-To-SQL

Reasoning Robustness of LLMs to Adversarial Typographical Errors

2024-11-08 · Esther Gan, Yiran Zhao, Liying Cheng, Yancan Mao 외

Large Language Models (LLMs) have demonstrated impressive capabilities in reasoning using Chain-of-Thought (CoT) prompting. However, CoT can be biased by users' instruction. In this work, we study the reasoning robustnes…

GSM8K MMLU