Akshay (@akshay_pachaar)RAG 시스템의 검색 성능이 5천 개 문서에서는 90%였지만 50만 개 문서로 확장하자 50%로 급락하는 사례를 제시하며, 동일한 임베딩 모델과 리트리버를 써도 문서 규모 증가가 성능 저하를 유발할 수 있음을 짚는다. 대규모 RAG 설계의 핵심 문제를 묻는 LLM 인터뷰 질문이다.https://x.com/akshay_pachaar/status/2052371239520629243#rag #llm #embeddings #retrieval #nlp
Related
China’s YouTube, Bilibili, plans global expansion, English-language sitehttps://www.semafor.com/article/08/19/2026/china...
China’s YouTube, Bilibili, plans global expansion, English-language sitehttps://www.semafor.com/article/08/19/2026/chinas-youtube-bilibili-plans-global-expansion-english-language-s...
The long tail was a UI problem.Every feature you add complicates the product for every user who doesn't need it, so deve...
The long tail was a UI problem.Every feature you add complicates the product for every user who doesn't need it, so developers build for the top of the demand curve and the rest go...
Stripe、AIモデルゲートウェイのOpenRouterを買収https://www.watch.impress.co.jp/docs/news/2134389.html#watch_impress #テック #AI
Stripe、AIモデルゲートウェイのOpenRouterを買収https://www.watch.impress.co.jp/docs/news/2134389.html#watch_impress #テック #AI