Creating a Wordle Bot- Part 2 - Learnings
Since I am new to 4o and not an experienced coder, there were several things I learnt that may be self evident to seasoned coders but I am listing them in case they are useful to somebody in my shoes, i.e., somebody semi technical trying to use GPT to bridge those gaps.
Lesson 1: Build block by block: You cannot treat the code like a blackbox and assume GPT will figure it out. So it helps to build the code block by block, testing each function to ensure that it works and checking edge cases before moving on to the next block rather than building the entire thing in one go
Lesson 2: Be explicit in the instructions: When the logic gets complicated, it helps to spell out every step in the prompt so there is no scope for misunderstanding.
Lesson 3: Use examples in both instructions and testing output: For creating the match string logic, it really helped when I put out a few concrete examples and then asked 4o to print out a few examples of output generated from the code it had written to identify edge cases that were not handled well. It took a few tries to get the logic right.
Lesson 4: Some warnings can be ignored: When running the code in Pycharm, it always gave some warnings that I learned can make the code more elegant but can be ignored e.g., when you import libraries that are unused, or when you use similar sounding variables in the global context as well as inside a function or if two sections of code are similar looking, it can give a warning
Lesson 5: Sometimes you have to treat it as a blackbox: This is directly contradicting Lesson 1 but while handing large quantities of data, the processing can become time taking and python provides a multiprocessing library to parallelize the processing. While this is a concept I understand theoretically, it was a bit advanced for me to understand step by step. So while not ideal, I just treated it as a blackbox, let 4o handle it and whenever it threw an error, I pasted it back into 4o and let it deal with it until the code executed correctly. I intend to revisit it when I have time but since this is not my main focus, I am ok not understanding every nuance.
Lesson 6: Understanding data structures: One concept I could not get away from was understanding the data structures like decision trees and nested dictionaries to solve this problem. Understanding how data can be stored and how to process it is essential to the approach in some of these problems
Lesson 7: Understanding Computational Complexity: The Big O notation was something I came across while preparing for product manager interviews but the practical implications hit me when I hit execute and it was still going after several minutes. An additional learning was to ensure you print intermediate output snippets regularly to understand how far the execution has gone and which parts are taking time.
Lesson 8: Sometimes you need a fresh start: It is important to not fall in love with an approach. I had initially thought that the best way to do this was to use frequencies of letters but reached a dead end and instead I went back to the drawing board and tried using the number of words eliminated at each stage(an approach similar to what the NYT bot uses though there are differences in how it does it and what I finally settled on).
