Research15h ago
Attention Maps Poorly Predict VLM Reliability
A study of LLaVA-1.5, PaliGemma, and Qwen2-VL reveals that attention structure is a near-zero predictor of correctness. Researchers used a new VLM Reliability Probe to debunk the intuition that sharp attention maps imply calibrated answers. This finding warns practitioners that visual attention heatmaps are unreliable indicators of whether a model is hallucinating.