Here’s the latest installment in the series on working with LLMS: https://thenewstack.io/techniques-for-using-llms-to-help-write-a-postgres-function/
It was only after I deployed the function and used it in three different dashboards that the penny finally dropped. This had begun as an exercise to wrap a SQL idiom that made use of the lag() function. But when I looked at the final version of the function, I realized it wasn’t using lag(). Nor was it using dense_rank(). Just row_number() with partition logic was sufficient. Why, I asked ChatGPT, wasn’t that initially obvious?
ChatGPT: The final solution using ROW_NUMBER() and partition logic was not immediately apparent due to a combination of reliance on familiar patterns, the evolving understanding of requirements, and the iterative nature of problem-solving. This process highlights the importance of flexibility and openness to re-evaluating initial approaches to find the simplest and most effective solution.
It’s a good answer that synthesizes wisdom drawn from the human minds represented in the corpus of SQL queries and related conversations that ChatGPT feeds on.
The rest of the series:
1 When the rubber duck talks back
2 Radical just-in-time learning
3 Why LLM-assisted table transformation is a big deal
4 Using LLM-Assisted Coding to Write a Custom Template Function
5 Elevating the Conversation with LLM Assistants
6 How Large Language Models Assisted a Website Makeover
7 Should LLMs Write Marketing Copy?
8 Test-Driven Development with LLMs: Never Trust, Always Verify
9 Learning While Coding: How LLMs Teach You Implicitly
10 How LLMs Helped Me Build an ODBC Plugin for Steampipe
11 How to Use LLMs for Dynamic Documentation
12 Let’s talk: conversational software development
13 Using LLMs to Improve SQL Queries
14 Puzzling over the Postgres Query Planner with LLMs
15 7 Guiding Principles for Working with LLMs
16 Learn by Doing: How LLMs Should Reshape Education
17 How to Learn Unfamiliar Software Tools with ChatGPT
18 Using AI to Improve Bad Business Writing
19 Code in Context: How AI Can Help Improve Our Documentation
20 The Future of SQL: Conversational Hands-on Problem Solving
21 Pairing With AI: A Senior Developer’s Journey Building a Plugin
22 How LLMs Can Unite Analog Event Promotion and Digital Calendars
On the contrary, Jon, it seems to me a perfectly banal answer. I mean this is all I ever see from ChatGPT: bland, banal answers that sometimes land in the strike zone of reasonable bland, banal answers, and sometimes land outside it. Which I suppose is also what you get from most *people*, but that’s also why I don’t talk to most people unless I’m getting paid for it and need the money.
The robot told you that well, you had to figure it out along the way. But that does not answer your question. On the other hand, the robot does not know why you didn’t see that faster: only you do. The robot does not know why it didn’t figure it out faster because it’s only stringing proximal sounds together in ways it’s seen done before.
> that’s also why I don’t talk to most people
> it’s only stringing proximal sounds together in ways it’s seen done before
That’s one way to describe it: “just next-word prediction”
Another way: pattern-based search that finds things people have written related to the pattern.