I have to use a ton of regex in my new job (plz save me), and I use ChatGPT for all of it. My job would be 10x harder if it wasn’t for ChatGPT. It provides extremely detailed examples and warns you of situations where the regex may not perform as expected. Seriously, try it out.
I’ve tried it and found it wanting at regex and excel formulas, but I’m glad to hear it’s working for you! Are you using 4? I haven’t tried that one and I hear it’s better.
I typically try 3.5 first and switch to 4 if the results aren’t great. 3.5 typically handles basic use cases quite well, for example, writing regex that detects jira ticket naming nomenclature. For more complex things, I go to 4.
It sometimes gets things wrong, but I’ve also found that just saying “that didn’t work” gets it to reevaluate for more complex situations
it helps if you hold ChatGPTs hand and walk it through what you need. For example if you have a regex with 3 requirements, ask it to write a regex for the first requirement, then ask it to modify the previous output to add another requirement, and so on. that way you can sort of “audit” it as it generates the correct regex.
there is some more discussion of this in a similar post from a few days ago.
So I was trying to write a regex for use with my ChatGPT discord bot. I wanted to trim off any final partial sentence at the end. I went around and around with it for a couple of hours because look ahead and look behind are just not something I do often enough.
It kept writing more and more complicated regex that didn’t work. The final solution, while not exactly perfect - it won’t keep a quote at the end of a sentence, and honorifics like Mr. and Dr. throw it - it wasn’t nearly as complicated as ChatGPT was making it. It still never did give me anything working - I just fucked around on regex101 until I got it right. As usual but having wasted 90 minutes or so.
I’ve found that you need to be very careful when asking it to modify things it produced directly without making significant changes to the regex it provides. Once I get to the 3rd or 4th iteration of asking it to modify previous responses I’ve found the likelihood that it starts hallucinating to increase dramatically. The best solution I’ve found to this is to put your entire request in a single prompt that walks it through all requirements step-by-step.
You can improve the reliability if you provide it test cases. You can now be the PM you wish you had for the robot that will eventually replace you.
