Thank you to everyone who participated in this post and shared your thoughts about how to handle apps produced with LLM technology.

Any project posted here will now require an AI disclosure if any ML/LLM technology was used in any way in the development:

  • Please include [AI] or [AI Assisted] as a tag at the end of your post title when sharing or discussing the project.
  • In the post body, please include in what way AI was used, e.g.:
AI disclosure:
design: none
implementation: assist
testing: copilot
documentation: auto
review: pair
deployment: hint

Please consult ai-declaration.md for details how to apply about these categories. If you are posting about an app, but are not the developer, you should include the most complete information available, e.g.: " This repository lists Copilot as a contributor." or " This app shows signs of AI development such as…"

Why not ban AI entirely?

There were a few comments that called for an outright ban on any apps that were built in part or whole with LLMs. It was not a clear consensus. Furthermore, I don’t personally see how that would be feasible. Consider the following situations:

  • A dev uses an AI search tool to look up information related to the project that is incorporated into the project
  • A dev uses an AI tool to translate between languages
  • A dev creates all of the code from their brain without any use of an LLM, but uses an LLM for the project description/readme page
  • A dev used a modern code review to identify errors
  • A dev used AI to create icons or other assets for human written code
  • A dev incorporates code from another project, which used LLM tools in one or more of these ways
  • Etc.

I’m not sure which if any of these situations would lead to an app being banned from this community, or how to best navigate cases like this. Please DM me or reply here if you have a solution and would like to be in charge of implementing it as a moderator in this community.

  • aizakku@sh.itjust.works
    6·
    3 days ago

    This sounds great and if it’s able to be validated even better!

    I’m a software dev and historically tough problems took hours to days to solve. Sometimes you’d get lucky and find someone with the same issue in StackOverFlow and just “borrow” the code and adapt it to your use case. There is internal debate on doing things easy and fast, à la AI, or getting a dopamine rush from solving hard problems. I like solving hard problems but I also enjoy deploying and setting up apps on servers and playing with code and infrastructure so code isn’t the only way to solve problems. One day I might start a project where AI is the the dev, other times I want to roll up my sleeves and start typing my own code. So long story short,I haven’t figured out my stance with AI 100% myself.

    In my full time job, it’s a no brainer as I have multiple projects and lets me deliver faster and there’s a push from the top to use AI. In my personal dev experience I use local LLM with an open source VS Code plugin and have built full apps just to understand capabilities. I found out after using local models for a few months with a plugin that it basically doesn’t care if you’re signed in they’re siphoning chats locally to their servers. So I forked the repo with the source being in GitHub and had AI iterate over it for a day where it removed all the telemetry code. Got the vsix build for vs code, uninstalled the one from the marketplace and loaded my own build. It runs faster than before and gives me a piece of mind my data remains local. So AI good at doing grunt work like this, I don’t think I’d get much satisfaction of reading 100s of thousands of lines trying to identify all the different ways it was phoning home.

    The flip side is I’ve been developing a game on and off for a few years. I’ve handcoded everything and had a pretty dumb bug that I couldn’t figure out with the networking. I don’t want it to be AI and want it as a showcase for my own abilities. I wrote a queue for the game to hold networking events, and everything worked fine when running 3 or 4 instances on my local machine which stayed in sync basically because I could only interact with one instance at a time. So it seemed fine. I was using an array to use FIFO processing, but accidentally had a used pop_back instead of pop_front in the code. When I tried multiplayer with some fam and friends it epically failed as each client was woefully different. I read, re-read and tried to figure out what was wrong for days. Sometime you get blinded by looking at the code too much, and it was just that one spot I couldn’t see. After much mulling about I decided to use Qwen to take a look at the file, sure enough within seconds it found the bug 🤦‍♂️.

    So I’m not sure this story has a point, but hopefully the reader can gleam some insights to what a greybeard thinks about AI. It’s a powerful tool and has its uses. I see many people shut down and immediately lambast the poster (perhaps rightfully so) and froth at the mouth at even and eye batting something might be AI. It’s as if the reader sees AI is the beginning of the end of civilization. I mean I get that, but Pandora’s box has been open and unfortunately unless we get a Carrington event knocking out all electronic devices on the planet, it’s here to stay. We need to understand the people that use it and their intentions more than the output (though closed/open source software is another topic of discussion). I mean AI writes somethings better than me, other times it’s horrendous and goes down rabbit holes that make no sense.

    I understand AI is disruptive and there are many reasons for concern with security and reliability of AI generated slop code. I think any Lemmy client should be open source so folks can review the code and understand what they’re installing on their phone and fork it if needed. Disclosure of AI is paramount so each person can make their own informed consent with what software to engage with. Ultimately, newer generations won’t know anything different and AI designed apps will be as ubiquitous as anything else. I’m sure there will still be pockets of folks doing things they like, such as coding the human way, but the draw to AI is an interesting one and for myself I’ve not reached consensus. Maybe when I do, I’ll leave the field of software development as it’s an entirely different game today then it was a few years ago.

    Disclosure: no AI used for this post, just some dude trying typing on his phone.