Attention competition predicts context rot in retrieval
Filler tokens maintain accuracy up to 32K tokens in retrieval but rival statements multiply loss by 9-110 across five open models.
1 stories, newest first · one card per story
Filler tokens maintain accuracy up to 32K tokens in retrieval but rival statements multiply loss by 9-110 across five open models.
Today 1K readers · 1.5K views · 474 reads · all time 184K views