One thing I didn't have a problem with when I first got into home monitoring and control was the software changing underneath me. I fully realize that software changes over time and that we need to adapt if we want to keep up with the industry, but MONTHLY? REALLY??
Home Assistant has a monthly release schedule, and many of their releases cause changes in the various add-ons and integrations. Dashboards you spent a week getting "just right" change. New features don't work the way you expected. Plus, each of those custom items (especially cards) has its own updates that change the way they work. It makes one (me especially) reluctant to update anything once we get it working. But, for me, the real annoyance is changes to our toolset to create new things.
Espressif (the folks that created the ESP32 series) is especially harrowing to people like me that build their own stuff to control and monitor things. Their libraries change often, and code we compiled and worked with yesterday fails to compile when we update to get the latest new capability or bug fix. I had that happen too often. Then, you have to hunt for the release documentation and hope you find a clue to fixing the source that worked yesterday.
Between the tools and underlying system changing as we work on things and then waiting for bugs to be fixed, it's hard not to just stick with whatever works right now and simply ignore any changes because there's usually some teeth-grinding involved in using the 'latest' of anything. I've seen users complain that a lot more emphasis should be put on making things work than on piling feature on top of feature that most of us won't even use.
Such was the case with my little Xiao ESP32C6 Zigbee temperature and contact sensor that was actively measuring the air temperature and watching the septic float. I installed a new release of ESPHome, and suddenly my cool little CYDs wouldn't work. Seems they changed some names of things I used.
This was a royal pain chasing down which things changed to what. I just gave up and turned he task over to an AI. It prowled through my code, looked at the new errors and gave me the new names I needed to put it. That trick saved me a ton of time prowling through forums loaded with ads and chasing my tail to get what I needed.
Remember this trick when you run into the same problem.
However, I still hold with not letting it write code for me. Especially when you have something that involves a lot of details. What happens to me is that I tell it what I want to do, and it does well until we're chasing a bug; then the conversation gets so long that the AI starts to confuse what worked and what didn't. It seems to get trapped in a loop where it simply confuses itself and can't recover.
I'm all for using whatever help I can get putting something together, and would proudly use AI to overcome syntax errors I just can't see or get a new way to shift some bits around to meet some obscure spec, but be really careful doing it. In a couple of instances it told me to remove files that were causing problems. DON'T do that.
If you do want to see just how well it does with code, be sure to back up everything in sight just to be sure it doesn't convince you that this is the best way to go and you delete a week's work.
And, as some of you probably noticed, I'm anthropomorphizing it already. Currently, I consider it a quirky research assistant of indeterminate sex. It's a co-worker that makes mistakes and gets lost in detail but is great at gathering information from obscure sites and putting it in words that I can understand. I can see a real danger though.
It's really good at finding things and telling me where to look for answers; how will that threaten future workers that aggregate data and present it in things like newsletters of new ideas and processes? Will we need humans to organize ideas and specifications for use? There are a huge number of people out there that are preaching the future of AI, and the Terminator movies didn't help us base our fears, real or imagined, on reality. People are already claiming it helps them with the stock market.
That kind of thing is for people that are far better than I at predicting the future of a new technology; I just want to use it as it applies to me. It saved me lots of time making the changes I inherited from the ESPHome codebase "update."
No comments:
Post a Comment