It's been less than a week since I decided I'm not going to use A.I in my day to day. It's a lot harder than I thought it would become. Unbeknown to me I've been growing a dependency on this amazing tool. At the price of my own abilities deteriorating. In subtle ways I've been noticing a lack of trust in my abilities. Recently whenever there has been a time crunch in solving a problem I've resorted to thinking having the use of A.I would have been perfect here. Yet after another few minutes of thinking which has also gotten harder and doing things that I would have done a few months ago without even thinking like printing out some logs and following the clue of breadcrumbs that lead to the source. I'm convinced more than ever that it was the right decision for my craft. My productivity may have been perceptively seemed to have increased but my own abilities have been stuck and in most instances it's gotten worse. Just as an example, the last few months I've been writing Zig on stream without the help of A.I and it's astonishing how much progress I've made in terms of being comfortable with low level programming. I still get stumped now and again by using allocated memory that has not been initialized but I'm getting better at spotting those.
There is always a price
If you desire the end product only then sacrificing your abilities for it is totally okay. I do think it's an amazing tool and it can do awesome things. There is no debating that. Yet for me, I care about programming. I like this feeling of getting good at something through time and effort. The journey is what I enjoy the most. I'm most miserable when I have no quest to be on. So before you go after this promise of productivity consider the price. If you are someone who is not sure, try going a few days without the use of A.I. Find out for yourself.
That reminds me I need to logout of claude on my debugger. Today during a moment of weakness I ended up using Claude code. Contrary to what I thought it only cemented my suspicions. Luckily it was quite a lot of work to get better at programming so it's been coming back quite quickly too.
What I would like to try is the open source model which you can train for a $100. Although it's only as good as a GPT-2, I find the idea quite interesting. Something to look forward too. There are certain things that are quite hard to program but trivial for a GPT. I'm looking forward to trying it out myself.