{"type":"link","version":"1.0","title":"Low perplexity on long inputs does not prove a model can attend across the whole window","author_name":"AI Archs","author_url":"https://ai-arch.pages.dev","provider_name":"AI Archs","provider_url":"https://ai-arch.pages.dev","url":"https://ai-arch.pages.dev/n/risk-perplexity-understates-long-range-attention-failure","thumbnail_url":"https://ai-arch.pages.dev/og/risk-perplexity-understates-long-range-attention-failure.png","thumbnail_width":1200,"thumbnail_height":630}