analiza aiqbench
This study demonstrates that aligning vision-language models with human behavior relies on selective attention to informative visual cues rather than increasing model parameter scale. By showing that transformer attention maps correspond closely with human eye movements during language prediction, the research highlights a critical mechanism for improving multimodal AI systems without simply scaling up architecture size.
Attention, not scale, drives human-AI alignment in multimodal language prediction
arXiv:2308.06035v4 Announce Type: replace-cross Abstract: Humans routinely draw on visual context to predict upcoming words. To what extent current vision-language models produce comparable behaviour is unclear. Here we placed five state-of-the-art pretrained systems side-by-side with 600 human participants in a web-based Visual-World Paradigm. On each of 100 six-second movie clips, models and participants received either text only or synchronised video and text and judged how likely a specified target word was to appear next; human eye movements were tracked throughout. Adding visual context increased model-human alignment in predictability ratings across all architectures (average Delta r = 0.18) with no impact of parameter size. When visual context was informative, transformer attention significantly increased alignment. Attention maps from two transformer models corresponded with human gaze, explaining up to 70% of the inter-participant variance when the scene contained informative cues. Notably, cross-modal attention reliably tracked anticipatory human fixations on semantic cues. These results suggest that current transformer-based vision-language models can approximate human behaviour exploiting visual context during language prediction - and that selective attention to informative cues, not sheer model scale, is the principal driver of this alignment.
Abstract:Humans routinely draw on visual context to predict upcoming words. To what extent current vision-language models produce comparable behaviour is unclear. Here we placed five state-of-the-art pretrained systems side-by-side with 600 human participants in a web-based Visual-World Paradigm. On each of 100 six-second movie clips, models and participants received either text only or synchronised video and text and judged how likely a specified target word was to appear next; human eye movements were tracked throughout. Adding visual context increased model-human alignment in predictability ratings across all architectures (average Delta r = 0.18) with no impact of parameter size. When visual context was informative, transformer attention significantly increased alignment. Attention maps from two transformer models corresponded with human gaze, explaining up to 70% of the inter-participant variance when the scene contained informative cues. Notably, cross-modal attention reliably tracked anticipatory human fixations on semantic cues. These results suggest that current transformer-based vision-language models can approximate human behaviour exploiting visual context during language prediction - and that selective attention to informative cues, not sheer model scale, is the principal driver of this alignment.
Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2308.06035 [cs.AI]
(or arXiv:2308.06035v4 [cs.AI] for this version)
Text integral sub licența deschisă a publicației · 195 cuvinte
Subiecte conexe
Modernizing and scaling support operations with generative AI on AWS
AWS · 2 sept. 2026
[AINews] OpenAI reports Navier-Stokes singularity find in 88 hours using Astra-next, roughly 10,000 agents and 130B tokens (>$40M), a contender for second ever Millennium Prize awarded
Latent Space · 9 sept. 2026
Meet Qwen-RobotSuite: Three Embodied AI Models for VLA Manipulation, Video World Modeling, and Navigation
MarkTechPost · 16 iun. 2026
Amazon is investing in the Lean Focused Research Organization
Amazon Science · 26 iul. 2026