# Rethinking Ground Truth: A Case Study on Human Label Variation in MLLM Benchmarking

arXiv: https://arxiv.org/abs/2603.19744 · HTML: https://mlsanity.com/paper/2603.19744/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 19 of its 34 references that are in the map.
- First posted on arXiv 20 Mar 2026.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 19 of 19)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes](https://mlsanity.com/paper/2005.04790/) (2020)
- [MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans?](https://mlsanity.com/paper/2408.13257/) (2024)
- [MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding](https://mlsanity.com/paper/2406.14515/) (2024)
- [MultiVerS: Improving scientific claim verification with weak supervision and full-document context](https://mlsanity.com/paper/2112.01640/) (2021)
- [The 'Problem' of Human Label Variation: On Ground Truth in Data, Modeling and Evaluation](https://mlsanity.com/paper/2211.02570/) (2022)
- [Taming Overconfidence in LLMs: Reward Calibration in RLHF](https://mlsanity.com/paper/2410.09724/) (2024)
- [Exploring Hate Speech Detection in Multimodal Publications](https://mlsanity.com/paper/1910.03814/) (2019)
- [Measuring the Reliability of Hate Speech Annotations: The Case of the European Refugee Crisis](https://mlsanity.com/paper/1701.08118/) (2017)
- [On the Calibration of Large Language Models and Alignment](https://mlsanity.com/paper/2311.13240/) (2023)
- [VariErr NLI: Separating Annotation Error from Human Label Variation](https://mlsanity.com/paper/2403.01931/) (2024)
- [Detecting Hate Speech in Multi-modal Memes](https://mlsanity.com/paper/2012.14891/) (2020)
- [Two Contrasting Data Annotation Paradigms for Subjective NLP Tasks](https://mlsanity.com/paper/2112.07475/) (2021)
- [Detecting and Correcting Hate Speech in Multimodal Memes with Large Visual Language Model](https://mlsanity.com/paper/2311.06737/) (2023)
- [Calibration of Large Language Models on Code Summarization](https://mlsanity.com/paper/2404.19318/) (2024)
- [Rethinking Multimodal Content Moderation from an Asymmetric Angle with Mixed-modality](https://mlsanity.com/paper/2305.10547/) (2023)
- [AI vs. Human Moderators: A Comparative Evaluation of Multimodal LLMs in Content Moderation for Brand Safety](https://mlsanity.com/paper/2508.05527/) (2025)
- [PoliTok-DE: A Multimodal Dataset of Political TikToks and Deletions From Germany](https://mlsanity.com/paper/2509.15860/) (2025)

## Abstract (arXiv)

Human Label Variation (HLV), i.e. systematic differences among annotators' judgments, remains underexplored in benchmarks despite rapid progress in large language model (LLM) development. We address this gap by introducing an evaluation protocol for multimodal large language model (MLLM) benchmarking that explicitly accounts for two conditions: (1) human label agreement and (2) disagreement. We apply this protocol to two state-of-the-art MLLM families (Gemma 3, Qwen 2.5 VL) using non-aggregated human annotations from a social media content classification dataset. Across tasks, we find that larger models tend to perform best on high-agreement subsets, yet often underperform medium-sized models when human disagreement is high, indicating that parameter count alone does not determine sensitivity to ambiguity and subjectivity. These results show that benchmarks based solely on consensus labels can overstate model capabilities in such domains and that incorporating human label variation yields more realistic and robust assessments of MLLMs in content moderation pipelines.