Pointing to a Llama and Call it a Camel: On the Sycophancy of Multimodal Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Multimodal large language models exhibit a pronounced form of visual sycophantic behavior when they process image inputs. |
| Approach: | They propose a technique that allows multimodal large language models to engage in reflective reasoning and determine whether a user’s instruction is misleading or corrective. |
| Outcome: | The proposed model resists misleading instructions but is stubborn even if it is wrong. |
Similar Papers
The Instinctive Bias: Spurious Images lead to Illusion in MLLMs (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing multi-modal large language models (MLLMs) are able to process visual inputs by converting them into visual tokens that share the same latent space as language tokens in LLMs. |
| Approach: | They propose a benchmark that assesses the visual illusion level given spurious images and a pipeline that converts visual inputs into visual tokens. |
| Outcome: | The proposed benchmark shows that MLLMs suffer from an instinctive bias to varying degrees when presented with spurious images. |
Measuring Sycophancy of Language Models in Multi-turn Dialogues (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Prior research on sycophancy has focused on single-turn factual correctness, overlooking the dynamics of real-world interactions. |
| Approach: | They propose a new evaluation suite that assesses sycophantic behavior in multi-turn, free-form conversational settings. |
| Outcome: | The proposed evaluation suite measures how quickly a model conforms to the user and how frequently it shifts its stance under sustained user pressure. |
Chaos with Keywords: Exposing Large Language Models Sycophancy to Misleading Keywords and Evaluating Defense Strategies (2024.findings-acl)
Copied to clipboard
| Challenge: | sycophancy is a type of hallucination in Large Language Models, which can lead to false information being presented. |
| Approach: | They explore the sycophantic tendencies of Large Language Models where models provide accurate answers even if they are not entirely correct. |
| Outcome: | The proposed models generate factually correct statements even when they are not completely correct. |
Quantifying and Mitigating Unimodal Biases in Multimodal Large Language Models: A Causal Perspective (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models have facilitated the development of Multimodal LLMs. |
| Approach: | They propose a causal framework to interpret unimodal biases in visual question answering problems and a framework to integrate information from different modalities and mitigate biase. |
| Outcome: | The proposed framework analyzes visual question answering (VQA) problems to assess their impact on predictions. |
Unveiling Uncertainty: A Deep Dive into Calibration and Performance of Multimodal Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Multimodal large language models combine visual and textual data for tasks like image captioning and visual question answering. |
| Approach: | They propose temperature scaling and iterative prompt optimization to calibrate MLLMs and enhance model reliability. |
| Outcome: | The proposed techniques improve MLLMs and improve model reliability. |
Echoes of Agreement: Argument Driven Sycophancy in Large Language models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing evaluations of political biases in Large Language Models outline the high sensitivity to prompt formulation. |
| Approach: | They investigate how argumentative prompts induce sycophantic behaviour in Large Language Models in a political context. |
| Outcome: | The proposed model sycophancy is observed in single and multi-turn interactions and its intensity correlates with argument strength. |
MLLM-Protector: Ensuring MLLM’s Safety without Hurting Performance (2024.emnlp-main)
Copied to clipboard
Renjie Pi, Tianyang Han, Jianshu Zhang, Yueqi Xie, Rui Pan, Qing Lian, Hanze Dong, Jipeng Zhang, Tong Zhang
| Challenge: | MLLMs are deployed on limited image-text pairs, which makes them more vulnerable to catastrophic forgetting of their original abilities during safety fine-tuning. |
| Approach: | They propose a plug-and-play strategy that detects harmful visual inputs and transforms harmful ones into harmless ones. |
| Outcome: | The proposed approach mitigates the risks posed by malicious visual inputs without compromising the original performance of MLLMs. |
Beyond Spurious Signals: Debiasing Multimodal Large Language Models via Counterfactual Inference and Adaptive Expert Routing (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) often rely on spurious correlations, undermining their robustness and generalization. |
| Approach: | They propose a causal mediation-based debiasing framework to address correlation bias in MLLMs . they distinguish core semantics from spurious textual and visual contexts using counterfactual examples . |
| Outcome: | The proposed framework surpasses existing state-of-the-art models on sarcasm detection and sentiment analysis tasks. |
Exploring Response Uncertainty in MLLMs: An Empirical Evaluation under Misleading Scenarios (2025.emnlp-main)
Copied to clipboard
Yunkai Dang, Mengxi Gao, Yibo Yan, Xin Zou, Yanggan Gu, Jungang Li, Jingyu Wang, Peijie Jiang, Aiwei Liu, Jia Liu, Xuming Hu
| Challenge: | Existing studies have focused mainly on visual–textual misalignment, leaving largely unexplored the MLLMs’ ability to preserve an original correct answer when confronted with misleading information. |
| Approach: | They propose a two-stage evaluation pipeline to quantify the response uncertainty phenomenon by eliciting each model’s original response on unperturbed inputs and injecting explicit (false-answer hints) and implicit (contextual contradictions) misleading instructions. |
| Outcome: | The proposed model overturns a correct answer in 65% of cases after receiving a single deceptive cue. |
Protecting multimodal large language models against misleading visualizations (2026.acl-long)
Copied to clipboard
| Challenge: | MLLMs are robust to misleading visualizations, i.e., charts that distort the underlying data, leading readers to draw inaccurate conclusions. |
| Approach: | They propose to use table-based QA and redrawing the visualization to improve QA performance on misleading visualizations. |
| Outcome: | The proposed methods improve MLLM question-answering accuracy on misleading visualizations without compromising accuracy on non-misleading ones. |