Give the Model a Finger: Developmentally Aligned Evaluation of Counting in Vision-Language Models
Vision-language models (VLMs) are exact on small sets and drift into estimation on larger ones, a profile reminiscent of children before they master counting.
ProofPaper ↗
Key points
- Children cross the barrier with epistemic actions: pointing at each item while reciting number words, moving counted items aside, partitioning the array.
- We give VLMs the same actions as dumb image tools with no visual intelligence of their own in a training-free agent loop.
Sources (1)
- [1]Give the Model a Finger: Developmentally Aligned Evaluation of Counting in Vision-Language ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 07:19 AM
Vision-language models (VLMs) are exact on small sets and drift into estimation on larger ones, a profile reminiscent of children before they master counting.
Children cross the barrier with epistemic actions: pointing at each item while reciting number words, moving counted items aside, partitioning the array.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026How Language Models Organize Task Information
- Oct 19, 2026Guided Diffusion for Trustworthy Synthetic Text Generation
- Oct 19, 2026Target-Aware Curation of Hallucination Stress Tests for Large Language Models
- Oct 18, 2026From Seeing to Reporting: A Theory of Reliable Visual Set Generation in Vision-Language Models
- Oct 18, 2026Conformal Prediction-Enabled Sequence-level Policy Optimization for Reasoning Language Models
- Oct 17, 2026A Symmetry-Matching Approach to Blind-Spot Reduction in Sparse Autoencoders