Visual memory raises recall but lowers precision in long video
On VMem-Bench, retrieval lifted entity recall from 0.559 to 0.628-0.693 while precision and forbidden-entity avoidance fell.
3 stories, newest first · one card per story
On VMem-Bench, retrieval lifted entity recall from 0.559 to 0.628-0.693 while precision and forbidden-entity avoidance fell.
VISTA's visual-memory harness lifts Claude Opus 5.0 to a perfect ARC-AGI-3 score using 57.4% fewer actions.
KeyRec lets VLMs answer long-video questions using 10% of dense visual tokens via bounded memory.
Today 380 readers · 425 views · 78 reads · all time 202K views