Rag · AI Beat

03 Sep 2026 · AI Beat Desk

Pages Nobody Was Meant to Read

Trellner Research tested Perplexity's Sonar models across 380 software categories and found three newly-registered domains had published 215,128 machine-generated "best software" pages — HTML-titled "Facts & Grounding Page" — that Perplexity cites extensively. 59.8% of its citations lead to domains outside the global top-100,000.

31 Mar 2026 · AI Beat Desk

Microsoft's Harrier Embeds 32K Tokens at Once

Microsoft released Harrier-OSS-v1, a family of decoder-only multilingual embedding models (270M, 0.6B, 27B) with a 32,768-token context window — roughly 30–60x longer than the 512–1,024 token ceiling most practitioners hit today. The 27B model takes SOTA on Multilingual MTEB v2 at 74.3; all three variants are MIT licensed.