mlsanityData to 7 Oct 2026

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

arXiv 2603.19744 ↗First posted 20 Mar 2026cs.CLcited by 0

On the map: LLM reasoning → Reasoning benchmarks and evaluation

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.

Key facts. No citing papers in the map yet (data through 7 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.

Builds on

19 of 34 references.

References · 19

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2023Improved Baselines with Visual Instruction Tuningcited by 4,362
  3. 2020The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memescited by 485
  4. 2024MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans?cited by 284
  5. 2024MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understandingcited by 180
  6. 2021MultiVerS: Improving scientific claim verification with weak supervision and full-document contextcited by 140
  7. 2022The 'Problem' of Human Label Variation: On Ground Truth in Data, Modeling and Evaluationcited by 133
  8. 2024Taming Overconfidence in LLMs: Reward Calibration in RLHFcited by 105
  9. 2019Exploring Hate Speech Detection in Multimodal Publicationscited by 103
  10. 2017Measuring the Reliability of Hate Speech Annotations: The Case of the European Refugee Crisiscited by 95
  11. 2023On the Calibration of Large Language Models and Alignmentcited by 72
  12. 2024VariErr NLI: Separating Annotation Error from Human Label Variationcited by 42
7 more
  1. 2020Detecting Hate Speech in Multi-modal Memescited by 38
  2. 2021Two Contrasting Data Annotation Paradigms for Subjective NLP Taskscited by 12
  3. 2023Detecting and Correcting Hate Speech in Multimodal Memes with Large Visual Language Modelcited by 12
  4. 2024Calibration of Large Language Models on Code Summarizationcited by 12
  5. 2023Rethinking Multimodal Content Moderation from an Asymmetric Angle with Mixed-modalitycited by 9
  6. 2025AI vs. Human Moderators: A Comparative Evaluation of Multimodal LLMs in Content Moderation for Brand Safetycited by 8
  7. 2025PoliTok-DE: A Multimodal Dataset of Political TikToks and Deletions From Germanycited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Threading the Needle: Reweaving Chain-of-Thought Reasoning to Explain Human Label Variation
  2. 2024MM-Soc: Benchmarking Multimodal Large Language Models in Social Media Platforms
  3. 2026Human Label Variation as Stable Signal: Learning Annotator-Specific Explanation Behavior via Cross-Annotator Preference Optimization
  4. 2025Maximizing Signal in Human-Model Preference Alignment
  5. 2023Understanding and Predicting Human Label Variation in Natural Language Inference through Explanation
  6. 2025ACT as Human: Multimodal Large Language Model Data Annotation with Critical Thinking
  7. 2023SEED-Bench-2: Benchmarking Multimodal Large Language Models
  8. 2024Assessment of Multimodal Large Language Models in Alignment with Human Values