Study finds multimodal training shapes text-only reasoning in language models
A new arXiv paper reports that multimodal training induces 'vision-shaped thinking' in language models even when they process text-only input, based on causal ablations across 10 models.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-13 05:00 · arXiv cs.AI
Emergent Vision-Shaped Thinking in Text-Only Reasoning through Multimodal Training