Comment by asah | Hacker News Reader

Comment by asah | Hacker News Reader

s/SQL/advanced SQL/ FTFY - SELECT/FROM/WHERE/GROUP BY for int/float/string is not hard or niche.

2c as DBA & researcher: advanced SQL is becoming irrelevant because LLMs are better at writing complex SQL than 99% of the available experts, who routinely botch NULL handling, datatype and index selection, performance optimization, etc. I see no reason why anybody should memorize the arcane syntax of SQL features like sliding window functions, exception handling in stored procedures, etc.

If this seems radical, consider that experts rely on query optimizers and EXPLAIN and people rely on the planner/optimizer for 99+% of queries. That was a radical position in the early 1980s.

I would love it if an AI could build what I’ve been banging my head against for over a year. How many lines of SQL can an LLM write and does it come up with the pipeline architecture all on its own? How does it incorporate subject matter experience to aggregate or transform the data? Like I said, I would love it if I didn’t have to worry about the nitty gritty stuff. But I have not seen that type of performance out of an LLM. Also I don’t get the comparison between LLMs and query optimizers which seem to be much more deterministic and logical about how they choose to build the query plan.

Copy paste the CREATE TABLE statements then ask it to write the query. It's amazing.

I wouldn't even know how to prompt the complex queries I have in mind. For simple queries that an ORM could write, I see, but for something complex that generates actual data from the tables, I don't see it coming.

Would love to know which LLMs you’re using because every single one I’ve used has been hot garbage that can’t do anything beyond joining a couple tables and fails miserably at basic GROUP BY/COUNT aggregates. Window functions? Fuggetaboutit

Paste an example and I'll show you! For me, I use ChatGPT 4 and start with the create table statements.

Reply on news.ycombinator.com