The last 2 weeks we were flooded with posts about DeepSeek. I am not writing about DS, but it is related, I cant escape 🙂
This article is excellent to sharp your skills as an engineer. This is even more important as the AI world is moving rapidly towards an Open Source world, where we will have a plethora of options available.
Uber has introduced QueryGPT, a tool that leverages generative AI to translate natural language prompts into SQL queries, streamlining data analysis processes. With LLM’s, QueryGPT reduces query authoring time from approximately 10 minutes to 3 minutes. It is cool and impressive, what I found interesting are some restrictions and assumptions:
– “We introduced “workspaces,” which are curated collections of SQL samples and tables tailored to specific business domains”
– “Another issue we found was that it’s incredibly challenging to go from a user’s natural language prompt to finding the relevant schemas”
– “Allowing users to select the tables used in the query generation came up as feedback from some users who saw that the tables that were eventually picked by QueryGPT were not correct “
– “We’ve been experimenting with prompts to reduce hallucinations”
So, we still need to understand the problem space and come up with solutions tailored to our business, let alone don’t forget it will always hallucinate at some point.