ChatGPT is everywhere these days.
OR folks seem not to talk about it too much, or at least not in the context of daily work.
Here are six ways ChatGPT can save you loads of time daily. On top, I will also present three other tempting ones, where I would not trust ChatGPT too much.
I grouped the use cases by three areas we need to combine daily – software engineering, dealing with people and mathematical model building.
Code engineering
1. Code templates
ChatGPT can come up with initial Python implementations of the most prevalent problem types:
Looks reasonable. I could imagine using this as a start to formulating my own problem.
2. Debugging
ChatGPT appears to be able to debug the code which it receives.
Disclaimer: The example I used originates from this repository.
Use this at your own risk though. There might be some bugs remaining!
3. Understanding pre-existing code
ChatGPT can give you an explanation of the code you found someplace on the Internet or which was developed by an old colleague:
This explanation is reasonable, and perhaps if prompted more, we could get a more detailed description.
Disclaimer: This is a simple problem formulation I took from the webpage of Gurobi.
Dealing with people
1. Explain the complicated technical concept to a non-technical audience
Good as a first draft. I would still change a few things, but at least the pain of coming up with v1 of the text can safely be outsourced to ChatGPT.
2. Write a follow-up email to ask for data
Just perfect. Put in your name and hit send.
3. Write an ad for your talk at the next meetup / conference
Again, a good first version but I wouldn’t send it yet.
Disclaimer: The title of the talk (Anything you can do, I can do better than you) is borrowed from the famous talk of Prof. Marco Lübbecke from RWTH, which he gave at one of the TEDx conferences. If you haven’t seen it yet, I highly recommend it!
Mathematical model building (DON’T do it)
… or do it at your own risk.
Some folks praised ChatGPT for being helpful when it comes to explaining mathematical concepts. Some even called it a tutor who is available 24/7.
So I gave it a try.
1. (Unsafe) Get a refresher on the basic concepts.
Umm… there’s nothing wrong here, but I would want to get more. I’d rather check with a textbook here.
Here ChatGPT behaves like a student who attended a lecture but learned everything by heart rather than trying to understand.
It could be helpful to have such a student handy every now and then, but I wouldn’t count on their support in anything beyond definitions.
2. (NO-GO) Get examples to help you understand a concept
Mmm.. that’s bad. Many things went wrong here. For example, I can give a better feasible solution by just looking at the problem.
I wouldn’t want a tutor to give me wrong explanations. So for now I would advise against asking ChatGPT to explain algorithms on examples.
3. (Use cautiously) Compare and contrast two modelling approaches.
Not bad, but – as I wrote earlier – I would be happy to see a more detailed answer.
Conclusion
ChatGPT can be super helpful with some things. (Un)fortunately, the most complex (and fascinating) part of our jobs as OR researchers appears to be beyond its reach – for now.
Stay tuned for future posts!