External LLMs are great for getting ideas and a quick overview of something, and helpers integrated into IDEs are useful to autocomplete longer lines of code or repetitive things.
I frequently ask ChatGPT to make whole functions for me. It’s important to check the code and test it, obviously, but it has saved me quite a bit of time.
I find it difficult to describe single functions that need to be integrated into a larger project. Especially if it needs to utilize a private or more unknown library. For instance, it totally fucked up using Bluetooth via DBus in C++. And the whole project is basically just that.
It certainly has its limitations. I’ve noticed a few topics where it generally gets things wrong, or I can’t seem to explain it properly. In that case, you may just use it as a reference guide. Maybe toss it some code and ask it what it thinks. It’s not always useful information, but sometimes that leads you down a different road that you would not have thought of before.
Problem is, I only ever need to use something more powerful than a search engine with topics that are too complicated for me and/or not well documented, in which case LLMs fail just as bad. So it’s actually only ever useful to get a general direction of a topic, but even then it could be biased to outdated information (eg. preferring bluetooth.h over DBus based bluetooth handling) or it outright doesn’t know new standards, libraries and styles. And in my experience, problems that have one, well accepted and documented standard don’t need any AI to get knowledge of.
External LLMs are great for getting ideas and a quick overview of something, and helpers integrated into IDEs are useful to autocomplete longer lines of code or repetitive things.
I frequently ask ChatGPT to make whole functions for me. It’s important to check the code and test it, obviously, but it has saved me quite a bit of time.
I find it difficult to describe single functions that need to be integrated into a larger project. Especially if it needs to utilize a private or more unknown library. For instance, it totally fucked up using Bluetooth via DBus in C++. And the whole project is basically just that.
It certainly has its limitations. I’ve noticed a few topics where it generally gets things wrong, or I can’t seem to explain it properly. In that case, you may just use it as a reference guide. Maybe toss it some code and ask it what it thinks. It’s not always useful information, but sometimes that leads you down a different road that you would not have thought of before.
Problem is, I only ever need to use something more powerful than a search engine with topics that are too complicated for me and/or not well documented, in which case LLMs fail just as bad. So it’s actually only ever useful to get a general direction of a topic, but even then it could be biased to outdated information (eg. preferring bluetooth.h over DBus based bluetooth handling) or it outright doesn’t know new standards, libraries and styles. And in my experience, problems that have one, well accepted and documented standard don’t need any AI to get knowledge of.