I'm seeing numerous surveys that most people are now "using AI" in their everyday work. I *think* this means generative AI, but I'm not sure.
1) Are you? How?
2) Should I be using it, and teaching my undergraduate students to do so?
3) Is my resistance just a later-generation version of "calculators will make us stupid", or am I correctly smelling an enshittification trap? It's not gonna stay free once we're all dependent on it.
@LouisIngenthron How do you get ChatGPT "customized to your coding environment"?
Most of my time with Unity seems to be spent researching the undocumented special feature for the task at hand, often interacting with forums because I can't find an answer by searching. Will AI help with this?
@peterdrake Yes, most likely.
I'm not sure if they have it for VS yet, but in VS Code, they have an extension called "CodeGPT". This plugin requires the paid API access to ChatGPT (they offer a free demo credit). With it, you get a normal ChatGPT window in your IDE. You can highlight a piece of code and ask it "what am I doing wrong here" and it'll have the necessary context to reply. You can ask it to rewrite or refactor your code. You can ask it to write a sample method in your language or explain how a function works. If there's enough data in its training set (which for Unity is probably a yes), it can be very helpful.
@peterdrake 1) Yes. I use GitHub CoPilot and I sometimes use ChatGPT in my work as a developer. The former is great at completing tedious simple jobs for me. The latter is good at summarizing thousands of stackoverflow questions and API documentation websites into useful answers, customized to my coding environment.
2) You should definitely be using it. Surely, there's something in your workflow that is repetitive and not terribly important that you'd prefer to offload to a machine?
As for your students, they do need to know about it but... the generative AIs are just as likely to reinforce bad habits as good, and they do occasionally just make stuff up. So, I would never recommend them to someone who is trying to learn a field. However, if you have the domain experience to know when they're wrong, they can act as a tremendous productivity booster.
3) A bit of both. Many companies are currently trying to stuff a square AI peg into a round hole, but where the technology does apply, it really excels.
Right now, they're basically running the same model as Gmail, where they offer it for free for personal use and make their money on the business customers. I don't see any reason they couldn't do that indefinitely, so if you just want it in that smaller personal capacity, it's pretty reasonable to expect to get it for free for some time.
Plus, there's a huge push to make this stuff open source. I've built and run LLMs on my local machine. It's a bit of a pain in the ass now, but it'll get easier. Definitely not up to the current ChatGPT standards, but it was able to do some pretty cool stuff nonetheless.
And custom-training LLM models is one of those easy-to-learn-but-difficult-to-master things, so anyone can get their feet wet with free software.