---
title: "Discussion on \"RAG's top-k retrieval has a black box problem, and it isn't fixed by a bigger model\" | Hashnode"
description: "Discussion on \"RAG's top-k retrieval has a black box problem, and it isn't fixed by a bigger model\". A vector search returns the nearest chunks by embedding distance, hands them to the model, and stops.\nNobody asks why those chunks and not the ones one rank lower.\nWhen retrieval brings back the wrong"
url: "https://hashnode.com/posts/rag-s-top-k-retrieval-has-a-black-box-problem-and-it-isn-t-fixed-by-a-bigger-model/6a782c5334ba1b96023037c6"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [RAG's top-k retrieval has a black box problem, and it isn't fixed by a bigger model](https://blog.bhuveshdhiman.com/rag-s-top-k-retrieval-has-a-black-box-problem-and-it-isn-t-fixed-by-a-bigger-model?utm_source=hashnode&utm_medium=feed)

A vector search returns the nearest chunks by embedding distance, hands them to the model, and stops. Nobody asks why those chunks and not the ones one rank lower. When retrieval brings back the wrong

Aug 9 · 1 min read · blog.bhuveshdhiman.com  
Read the article

-   [#rag](https://hashnode.com/tag/rag)
-   [#ai](https://hashnode.com/tag/ai)
-   [#retrieval-augmented-generation](https://hashnode.com/tag/retrieval-augmented-generation)

## Written by

[Bhuvesh Dhiman](https://hashnode.com/@bhuveshdhiman)

@bhuveshdhiman · Mumbai, India

  
[View profile](https://hashnode.com/@bhuveshdhiman)

Sharing learnings as a senior software engineer

## Responses

Markdown supported · ⌘↵ to post

No responses yet.