The Dark Side No One Talks About

[00:00:00]This is definitely a double-edged sword.

[00:00:03]Stick around and we'll [music] get right to it. Hey, welcome back guys. Jason KM4ACK.

[00:00:08]I wanted to put out basically an opinion video today. So, if you're not interested in my opinion, well, that's perfectly fine. Just kill the video right here and we'll call it even.

[00:00:20]But, for everyone else, what I want to talk about today is AI coding, also known as vibe coding.

[00:00:30]And in case you're not familiar, basically what that is is someone can sit down with an AI tool like ChatGPT or Claude and start writing code when they know absolutely nothing about code.

[00:00:46]And I've actually been doing some of this myself.

[00:00:50]But, there's some caveats to this and I kind of wanted to flesh those out a little bit today because it is definitely a double-edged sword. On one hand, it allows people that know nothing about coding to sit down and create something that they've got in their head.

[00:01:10]On the other hand, it allows somebody that knows absolutely nothing about coding to sit down and take something in their head and put it into ones and zeros. And lately, my inbox has been absolutely flooded with projects that people are starting to take on.

[00:01:29]And honestly, I just don't have time to review all of those.

[00:01:34]Um and I I don't want to offend anyone by no stretch of the imagination, but at the same time, I just can't do a deep dive on every single one that comes across my desk right now because there's literally so many of them coming into my inbox. I've had days where I'll get three or four different projects land in my inbox. People that they're excited about it and I understand that, but I just don't have time to adequately uh review each and every one of those that lands.

[00:02:07]All right, a slight interruption there with some other people passing me on the trail, but uh where was I? I think we were talking about all of the projects that landed in my inbox. And guys, like I said, I just don't want to offend anybody, but I literally I'm one person and I just don't have time because in order to really evaluate a project, it takes time to not only install it, but to play with it, to figure out its quirks, to find bugs that might be in it, and to give some honest feedback on that. And that can take multiple days when you really start diving into a project. Let's talk about uh one though in particular is Greywolf, and I'm not positive that it is fully AI-coded, but I know that the guy that's developing that has used Claude to put a lot of work into that.

[00:02:56]And by no means am I calling out the developer of Greywolf in any shape, form, or fashion. I don't know the guy, I don't know what he knows, I don't know what he doesn't know. Um I'm just giving Greywolf as an example because that is one project that I have had time to dive into, and he has done a remarkable job in a very short period of time. And the one thing that I like to see is he's responsive uh responsive to feedback, to bugs we might me find in the code, and he is rapidly developing, so those usually get fixed pretty quickly. But like I said, I've been working on my own project, and that's where some of these uh things came up at that I was noticing when I was using AI to write the code.

[00:03:42]Now, I chose to um well, let me back up. First of all, my project is a fork of someone else's.

[00:03:51]Uh there was just some things in there that I didn't completely agree with and I thought could be better. So, I took their code, dropped it into AI, and asked it to help me rewrite it. And between myself and AI, we chose to use Python as the underlying language of uh this application. And let me be right up front, I couldn't write out hello world in Python. I know nothing about it. My language that I wrote Build-a-Pi in and Sim 3 Linux was bash, and I'm pretty confident in my abilities in bash. I'm no uh expert developer, but I know enough to get by when it comes to the bash language. But, I don't think there was any way possible to do what I wanted to do inside of bash, and that's why we chose something like Python. But, as I went through this, I began to notice certain things. Now, let me uh just say right here that I am using a paid version of Claude AI.

[00:04:50]I'm not even running Claude code, I'm just running it in the browser.

[00:04:55]Uh so, I know there's some disadvantages to that, but that's the way I chose to work with this particular project. And overall, Claude has been really, really good to work with.

[00:05:06]But, there are sometimes that I noticed it just missed, and that's kind of where I want to hit on today's topic.

[00:05:14]You probably shouldn't just sit down with AI and try to write an application if you know nothing about the basics of the underlying fundamentals. So, let me give you an example here. One of the things I've Well, let me say this. Of course, this application is based around APRS.

[00:05:35]And that's something that I'm fairly familiar with. Not an expert by no stretch of the imagination, but I play with APRS on quite a frequent basis. And because of that, I was able to spot things as Claude was writing this code and I was testing things, I was able to spot mistakes that Claude just didn't pick up on before writing the code. One example in this is I was having I wanted the ability to send a bulletin from inside this application.

[00:06:09]And when Claude wrote the code to send a bulletin, I turned around and went to test it. And while it sent out something that appeared to be a bulletin, I also noticed a quirk with it. Normally, when a bulletin goes out, it will go out multiple times until that bulletin is killed. So, the purpose of a bulletin is something that lives quite a period of time and it gets resent every so often, say every 10 minutes or so. However, on a radio, that bulletin should only display one time.

[00:06:44]After that, the radio sees that it's a duplicate object and it will only display at that one time and then every other instance it drops. However, when I sent it from my new application, what I was seeing is every time the bulletin went out was transmitted, then it was showing up as a new bulletin on my radio and I knew that that wasn't the right behavior, so I started to dig into that a little bit. I used Direwolf so I could decode the raw packet that it was actually sending and what I found was Claude was putting a message ID on every single bulletin that went out and every time it went out, it was putting a unique message ID on it. And that's what was causing it to show up multiple times on my radio. Now, once I saw this, I told Claude, "Hey, you're not supposed to be using a message ID at all on a bulletin and you're certainly not supposed to be using a new message ID every single time a bulletin gets sent out because that's what was creating the issue. Once I told Claude this, it went ahead and looked up the APRS spec, verified what I was telling it was correct, and then corrected my code, and today it's working correctly. But I tell you all of this to say, you need to know something about the fundamentals of the application that you're trying to write. Now, what I mean by that is not the code itself, but in this case, I needed to know something about APRS so that I could give Claude uh good feedback in order to correct the mistake that it made. So, if I was going to try to write an application, let's say antenna modeling, that is something that I know nothing about. And chances are, me writing that application would uh be a train wreck, an absolute disaster. Uh you'd probably get uh both garbage in and garbage out at the exact same time. Now, on the other hand, I don't want to discourage you from getting in there and trying something like ChatGPT or Claude. I have found in my limited experience, and guys, I have very limited experience with this, but I have found with my limited experience that Claude is a better coding tool than ChatGPT or Gemini, that's the other two that I have tried right now. But I still would encourage you, I don't want to discourage you. I would encourage you to go ahead and try to interact with one of these AI agents, and take that idea that's floating around in your head, and try to put it to actual code.

[00:09:24]Uh at the same time, you may be super excited about the project, but you're going to have to see how many other folks are excited about the project, and you've got to gauge how much time do you want to spend on something that other people frankly just aren't super interested in. Now, for this current application that I'm working on, I look at it slightly differently because I am well aware that I'm going down the niche of a niche of a niche in APRS. And the application that I am writing, I might find three other people that subscribe to this channel that would even be remotely interested in what I'm building. However, I see this as an opportunity for myself to get better at interacting with an AI agent and create something that personally I want. So, I don't see it as a lost cause even if no one else is interested in this particular project. I'm certainly not going to get MRD interested in this project, and that is absolutely okay.

[00:10:32]Like I said, I'm doing this for me. I want to better my skills when it comes to this. I want to learn how to interact, how I can interact better, and how I can improve and get things done faster using an AI like Claude. So, if you've got an idea, then go ahead. Definitely give it a try.

[00:10:53]Just understand that AI will make mistakes, and the more you know about the underlying fundamentals, the better your application is going to be. And the other word of caution I will give you because I have been developing Build a Pi in 73 Linux for, I don't know, 8 years now or so, roughly.

[00:11:14]If you put it out in the wild and you allow other people to download it, well, they're going to find bugs. They're going to want things that you may not have intended with the original application, and you've got to decide how willing you are to support that going forward. So, hopefully this gives you a bit of insight before you start trying to create an application that you've had in your head for years using something like Claude.

[00:11:42]If you know a little bit about the fundamentals, then you're going to be fine. Just go in there and start playing with it, start poking at it, and you will get better at it. And remember this, the AI that you're using today is the worst version of AI you will ever use. It's going to be better in 2 months. It's going to be a whole lot better in 2 years. If you found today's information helpful, be sure to give us a thumbs up before you head off. We will see you guys on the next one. Until then, 73.