1) I needed to display a very high resolution image to a user. I have experience in GIS and imagery, so I knew I should use an image pyramid in some way, and from previous experience assumed I needed some server to cleverly serve the tiles. But I didn't want to implement it myself, and googling 'map server' led to rabbit holes. I consulted with chatGPT, and while it gave me several fake solutions, eventually it suggested using gdal2tiles.py to create the pyramid and then serve it directly with a CDN. This never occurred to me, and is a much better fit to this problem. This saved a LOT of time (on either building my own server or fudging with other solutions)
2) I have a streamlit service, and needed to use some of my infra inside, infra that was written using async-await. Unfortunately, streamlit and asyncio don't play nice together (boo streamlit). I went to chatGPT hoping it would find me a way to make it work anyway, and after trying everything it suggested (which failed), I tried googling myself, and spent several days without a solution. Eventually, I went back to chatGPT and it suggested building a small HTTP service that would serve the results, and access it in streamlit using requests, no async-await required. It's a hacky solution, but significantly faster than reimplementing my streamlit dashboard in another framework or rewriting my infra without async-await. It saved loads of time.
I think you'd agree these aren't junior level issues. ChatGPT definitely didn't solve every problem I came to it with, probably not even most, and even when it did I had to intervene signficantly. I feel the more experienced you are as a developer, the less valuable it is. But when you need to tinker in a field you're not proficient at, or if you need to brainstorm a solution to a tricky problem, it can be a great tool. I understand why many swear by it. It takes a while to learn what sort of issues are good to turn to chatGPT with and which aren't, and also how to phrase those issues
1) I needed to display a very high resolution image to a user. I have experience in GIS and imagery, so I knew I should use an image pyramid in some way, and from previous experience assumed I needed some server to cleverly serve the tiles. But I didn't want to implement it myself, and googling 'map server' led to rabbit holes. I consulted with chatGPT, and while it gave me several fake solutions, eventually it suggested using gdal2tiles.py to create the pyramid and then serve it directly with a CDN. This never occurred to me, and is a much better fit to this problem. This saved a LOT of time (on either building my own server or fudging with other solutions) 2) I have a streamlit service, and needed to use some of my infra inside, infra that was written using async-await. Unfortunately, streamlit and asyncio don't play nice together (boo streamlit). I went to chatGPT hoping it would find me a way to make it work anyway, and after trying everything it suggested (which failed), I tried googling myself, and spent several days without a solution. Eventually, I went back to chatGPT and it suggested building a small HTTP service that would serve the results, and access it in streamlit using requests, no async-await required. It's a hacky solution, but significantly faster than reimplementing my streamlit dashboard in another framework or rewriting my infra without async-await. It saved loads of time.
I think you'd agree these aren't junior level issues. ChatGPT definitely didn't solve every problem I came to it with, probably not even most, and even when it did I had to intervene signficantly. I feel the more experienced you are as a developer, the less valuable it is. But when you need to tinker in a field you're not proficient at, or if you need to brainstorm a solution to a tricky problem, it can be a great tool. I understand why many swear by it. It takes a while to learn what sort of issues are good to turn to chatGPT with and which aren't, and also how to phrase those issues