Is prompt engineering going to be the next "big thing"?

Very related fun fact: SQL was meant for business people so they can quickly generate reports w/out needing to rely on programmers. (When was the last time your CEO did a DB query on prod I wonder?!)

I wouldn't even trust them with a read only replica 😅

That bit of SQL trivia is hilarious 🤣

Fun fact COBOL was also marketed with claims to monitor or make programmers obsolete. It's designed for business and readable like plain English!

The very fact that you asked this makes me lose hope for the future of software.

"When did your CEO run a DB query in prod" - There's so much to unpack here I feel like I moved houses.

I don't think prompt will replace sql. SQL is minimal as is. You'll have to specify fields you want to select and filter on in your prompt which is almost an sql query. So it's pointless

If one can write intricate prompt to the point of getting SQL query he wants, he should probably simply write the SQL query from the start.

@daniel-wu I’m not so sure tbh. Yes, sql is the standard and was specifically designed to query data. But it’s also very technical and explicit. I wouldn’t exclude the possibility that there could be a way to query the data in a more concise way that is more like human language and has a lot of implied semantics.
Of course it introduces the problem of ambiguity but I think that this is not a fundamentally unsolvable problem.

I can imagine an interactive AI chat bot that you can ask to make a query and it does its best to infer what you want. And in case that it needs extra info from you, it can simply ask you about the specifics.

Also, SQL is designed for RELATIONAL data. This introduces a lot of technical complexity with JOINs and what not. If your data was hierarchical or a graph, there could be other, potentially simpler methods of querying it. See graphQL for example.

I don't think so. I'm sure it's going to be a side-task for many software engineers soon, but it will never be "big thing". Honestly the better NLP gets the less we need to "engineer" our prompts... So just by definition of NLP systems, prompt engineering as a job position is doomed from the start. Maybe there will be a couple of sparks in some big tech companies who think they are cool like that, but probably not for long.

@retoor i get your sentiment and you are correct if you want to stay close to the data layer, but a prompt that is translated to sql can be much shorter and more human friendly and semantically different. For example

Prompt: How much did we sell last year

SQL: SELECT SUM(order_total) FROM orders WHERE order_year = YEAR(CURRENT_DATE())

Notice that the word "sell" which has a semantic meaning doesnt appear in the SQL query and so its inferred by the prompt to provide the SQL command the appropriate context

@bioDan this is a great example for what I tried to say :)

@bioDan hmm, yes. But some programmer had to define what sell is and what filter criteria

@retoor depends on the infrastructure. Given enough training and testing data for prompts and the table schema, the LLM can predict with enough accuracy the user intention without explicitly defining what a "sell" is.

@retoor not necessarily. This can be inferred from the context.
AI/ML chatbots do this all the time.