From df24227bc87c103f7f0cfa3548408ac3d178f84c Mon Sep 17 00:00:00 2001 From: jannesDoege Date: Fri, 4 Oct 2024 12:34:56 +0200 Subject: [PATCH] Fixed typo in ai-sql-accuracy-2023-08-17.md The relevance search is depicted in a green box, not a red box. --- papers/ai-sql-accuracy-2023-08-17.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/papers/ai-sql-accuracy-2023-08-17.md b/papers/ai-sql-accuracy-2023-08-17.md index faf6a3b9..082b5edb 100644 --- a/papers/ai-sql-accuracy-2023-08-17.md +++ b/papers/ai-sql-accuracy-2023-08-17.md @@ -229,7 +229,7 @@ By providing just those 3 example queries, we see substantial improvements to th Enterprise data warehouses often contain 100s (or even 1000s) of tables, and an order of magnitude more queries that cover all the use cases within their organizations. Given the limited size of the context windows of modern LLMs, we can’t just shove all the prior queries and schema definitions into the prompt. -Our final approach to context is a more sophisticated ML approach - load embeddings of prior queries and the table schemas into a vector database, and only choose the most relevant queries / tables to the question asked. Here's a diagram of what we are doing - note the contextual relevance search in the red box - +Our final approach to context is a more sophisticated ML approach - load embeddings of prior queries and the table schemas into a vector database, and only choose the most relevant queries / tables to the question asked. Here's a diagram of what we are doing - note the contextual relevance search in the green box - ![](https://raw.githubusercontent.com/vanna-ai/vanna/main/papers/img/using-contextually-relevant-examples.png)