Aziz Mirzayev on Telegram: “RAG noto'g'ri javob bersa, ko'pincha LLM emas, r…”

Aziz Mirzayev @pro_aim · 938 subscribers

Post #1739 235 Oct 11, 2026, 10:03 UTC

RAG noto'g'ri javob bersa, ko'pincha LLM emas, retrieval aybdor ekan. O'xshash chunk degani kerakli chunk degani emas.

Bunga yechim sifatida vectorless RAG'ni taklif qilishyapti: hujjat bob → bo'lim daraxtiga aylanadi, agent esa shu daraxt bo'ylab yurib javobni topadi.

Men ham chunk'larga bo'lishda qiynalgandim. Endi shu yo'lini ham ko'rish kerak ekan.

Uzun hujjatlarni qanday bo'lyapsizlar: chunk'largami yoki strukturasi bo'yichami?

Manba: https://lnkd.in/p/dgd-WN3b

LinkedIn RAG Limitations and the Future of Retrieval | AI-Automation.in posted on the topic | LinkedIn 🚨 RAG Doesn't Always Fail Because the LLM Is Wrong. Sometimes the LLM fails because RAG retrieved the wrong information. The real problem may be retrieval, not generation. Traditional RAG : → Split documents into chunks → Create embeddings → Store them…

👍 1

Open in Telegram All posts from @pro_aim