Caption First, VQA Second: Knowledge Density, Not Task Format, Drives Multimodal Scaling

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zou, Hongjian, Ge, Yue, Ding, Qi, Liao, Yixuan, Chen, Xiaoxin
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913032253210624
author Zou, Hongjian
Ge, Yue
Ding, Qi
Liao, Yixuan
Chen, Xiaoxin
author_facet Zou, Hongjian
Ge, Yue
Ding, Qi
Liao, Yixuan
Chen, Xiaoxin
contents Multimodal large language models (MLLMs) have achieved rapid progress, yet their scaling behavior remains less clearly characterized and often less predictable than that of text-only LLMs. Increasing model size and task diversity often yields diminishing returns. In this work, we argue that the primary bottleneck in multimodal scaling is not task format, but knowledge density in training data. We first show that task-specific supervision such as Visual Question Answering (VQA) contributes little incremental semantic information beyond image captions: VQA signals can be reconstructed from captions with negligible performance loss. We then demonstrate that increasing knowledge density -- through structured caption enrichment and cross-modal knowledge injection -- leads to consistent performance improvements across multimodal and downstream benchmarks. Across controlled experiments, performance correlates more strongly with semantic coverage than with task diversity. These findings suggest that current MLLMs fail to scale primarily because training data lacks sufficient knowledge coverage. We advocate for knowledge-centric multimodal training as a principled foundation for scalable multimodal models.
format Preprint
id arxiv_https___arxiv_org_abs_2604_13054
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Caption First, VQA Second: Knowledge Density, Not Task Format, Drives Multimodal Scaling
Zou, Hongjian
Ge, Yue
Ding, Qi
Liao, Yixuan
Chen, Xiaoxin
Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
Multimodal large language models (MLLMs) have achieved rapid progress, yet their scaling behavior remains less clearly characterized and often less predictable than that of text-only LLMs. Increasing model size and task diversity often yields diminishing returns. In this work, we argue that the primary bottleneck in multimodal scaling is not task format, but knowledge density in training data. We first show that task-specific supervision such as Visual Question Answering (VQA) contributes little incremental semantic information beyond image captions: VQA signals can be reconstructed from captions with negligible performance loss. We then demonstrate that increasing knowledge density -- through structured caption enrichment and cross-modal knowledge injection -- leads to consistent performance improvements across multimodal and downstream benchmarks. Across controlled experiments, performance correlates more strongly with semantic coverage than with task diversity. These findings suggest that current MLLMs fail to scale primarily because training data lacks sufficient knowledge coverage. We advocate for knowledge-centric multimodal training as a principled foundation for scalable multimodal models.
title Caption First, VQA Second: Knowledge Density, Not Task Format, Drives Multimodal Scaling
topic Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.13054