I use them all the time. Recently I got ChatGPT to read a 50-100 page interface spec for a protocol (mostly structured prose as opposed to a typed interface spec) and skeleton out go client and server stubs. That probably would have taken me a day of copy/paste/check but with a few round trips I got acceptable results in under an hour.
Then I wanted to get some particular features out of a rust server that implements the protocol. I don't actually know Rust and the syntax is not superficially accessible to me, but with LLM I was able to one-shot translate a number of structs and functions and get a high-level idea of how stuff was working. I was able to port the features I wanted over pretty quickly. This was not entirely straightforward because async was involved on the Rust side but I wanted to use a more sync approach in the go client.
I also use it a lot to write out test stubs / skeletons. I paste in a bunch of interfaces or functions I want tests for and usually a pretty good starting point pops out. If it wasn't what I wanted I can refine interactively.
Overall I am very glad these tools exist now. Right now they're not doing anything for me I couldn't do myself but they help me maintain velocity and avoid getting stuck in gumption traps.
As someone also in a software career… spot on. I’d estimate GPT4 saves me 30-60 minutes every day, and I feel confident that’s not an exaggeration. It’s gotten to the point where I make certain scripts to automate things I previously would’ve done manually because the turn around with GPT4 for simple tasks means I’m spending 50% the time making an automated solution even for small tasks where it would normally take me too long to be reasonable to automate a one time task.
I use them all the time. Recently I got ChatGPT to read a 50-100 page interface spec for a protocol (mostly structured prose as opposed to a typed interface spec) and skeleton out go client and server stubs. That probably would have taken me a day of copy/paste/check but with a few round trips I got acceptable results in under an hour.
Then I wanted to get some particular features out of a rust server that implements the protocol. I don't actually know Rust and the syntax is not superficially accessible to me, but with LLM I was able to one-shot translate a number of structs and functions and get a high-level idea of how stuff was working. I was able to port the features I wanted over pretty quickly. This was not entirely straightforward because async was involved on the Rust side but I wanted to use a more sync approach in the go client.
I also use it a lot to write out test stubs / skeletons. I paste in a bunch of interfaces or functions I want tests for and usually a pretty good starting point pops out. If it wasn't what I wanted I can refine interactively.
Overall I am very glad these tools exist now. Right now they're not doing anything for me I couldn't do myself but they help me maintain velocity and avoid getting stuck in gumption traps.