Fetching from the wire…
01
02
03
04
05
06
07
08
Research2026-08-18 · source-backed
The authors formalize Dense Same-Class Attribute Misbinding and built InstaBind-Lite to measure it: 524 images, 529 groups of 3-6 same-class entities, 9,580 deterministically evaluated questions with source-instance annotations that separate copying from hallucination (arXiv 2608.16805). Five open-source models average 19.84% misbinding versus 7.55% for two commercial API systems, and ~81% of identifiable transfers come from adjacent instances. If you're running extraction over dense scenes, dashboards, or tables of similar items, standard accuracy metrics hide this entirely.
Each link below shares sources, entities, or timing with this story.
Shared entity: VLMs / Same source domain / Shared topic / Earlier coverage
Both cover VLMs; reported by the same outlet (arxiv.org); overlapping topics (accuracy, commercial).
Shared entity: Lite / Same source domain / Earlier coverage / Tension
Both cover Lite; reported by the same outlet (arxiv.org); earlier Lite coverage from 2026-07-26.
Shared entity: VLMs / Same source domain / Earlier coverage / Tension
Both cover VLMs; reported by the same outlet (arxiv.org); earlier VLMs coverage from 2026-06-29.
Both cover VLMs; reported by the same outlet (arxiv.org); earlier VLMs coverage from 2026-03-20.
Shared entity: VLMs / Same source domain / Earlier coverage
Both cover VLMs; reported by the same outlet (arxiv.org); earlier VLMs coverage from 2026-08-16.
Both cover VLMs; reported by the same outlet (arxiv.org); earlier VLMs coverage from 2026-08-16.
Both cover VLMs; reported by the same outlet (arxiv.org); earlier VLMs coverage from 2026-08-13.
Both cover VLMs; reported by the same outlet (arxiv.org); earlier VLMs coverage from 2026-07-30.