Well... I was watching hype about GLM5.2, so I went and asked it a question. Just a couple of years out of date. First time using it, first question to test its accuracy.
So much AI, so many add-ons... and so many bugs
- Thread starter PhineasD
- Start date
And you’d expect honest answers to question 3?
Not always
I have to admit, as much as I do like that we are seeing an influx of add-ons for XenForo, the fact that a lot of them are made using AI and it's obvious, and are released with so many bugs is a little worrying.
While I don't see XF's official Resources ever adopting it, I would suggest that any add-on using AI in any manner should be disclosed. That gives us the chance to avoid what could be potentially a sloppy and insecure add-on. If it's one of our long-time trusted developers who used it for some assistance, and they've reviewed the code, it's certainly better than this inrush of add-ons from developers who suddenly and recently surfaced after this forum has been around for at least 16 years.
I know some platforms are already requiring that disclosure (YouTube, etc.). It's not much of an ask to require that here as well.
I don't think AI is the problem, just the code quality, which is the coder / owner of it. There have been plenty of crappy addons, before AI. Maybe addons need to go through a QA process before going live on first upload? A few volunteers who just manage that. Is the code up to spec on creation, or not? Updates are not vetted. You can even have a bypass for known users who only upload quality addons? You're really just trying to stop the resource directory becoming an AI slop repository, not banning AI itself... and one check on creation can do that, then let the person go wild.
That crazy, I know plenty of talented software developers. I don't know any who don't use AI.
Many don't, either. I would trust existing, long-time XF add-on developers, though, as opposed to the newer slop from the unknowns.
The only addons i have are the official ones from Xenforo.
They're very good and they do the job for now.
It's not that i don't trust the others who make the other addons, i just don't think it's the right thing for me to be adding something that might look terrible and make my small group of members walk away from my forum.
I may look at using xon's sign up and abuse add on again because that too was a good one.
They're very good and they do the job for now.
It's not that i don't trust the others who make the other addons, i just don't think it's the right thing for me to be adding something that might look terrible and make my small group of members walk away from my forum.
I may look at using xon's sign up and abuse add on again because that too was a good one.
That is a good revenue source for XF IMO.... it would relatively quick for them to go through addon code purely to assess a handful of factors. They aren't doing it for right or wrong way of coding, just if it works in XF and isn't go to break things.
Obsidian.md has automated addon screening.
Their lying to you. In fact if they aren't using AI at all, don't employ them, unless of course your looking for something built with COBOL
I believe some agents are getting better with COBOL - although Gemini tells me that some of the older fixed-format COBOL code and legacy mainframe-specific stuff (z/OS, CICS, etc) still cause problems for the generic models that haven't been specifically trained on that environment.
I never did any real work with COBOL - only did a little bit of simple stuff at uni, so I don't know much about the environment, but my general feel is that there were no widely-implemented standards across platforms back then, so it's not a matter of "write COBOL once, and run it anywhere" like we have with languages like PHP. As such, my understanding is that the actual way to implement COBOL depends very much on the specific platform you're working with - which is probably why generic AI systems are not always great with doing real work with COBOL.
This is largely supposition on my part - like I said, I have no actual experience with it in a commercial setting.
That isn't a bad idea.XF should just offer a paid verification service where they will go through your code and verify it's good then you get a verified badge on the addon. Done.
The fees pay the person doing it.
Woltlab for examples downloads every theme, plugin and add on that is submitted in their marketplace. It receives an automated review based on things like the tar, bundled files, PIP, languages, templates, etc. Then a manual review happens by one of their developers and they carefully check the quality of the submitted work, along with the correct permissions, validations, classes, etc. They do take a 25% commission for this.
I agree! I am firmly for anything that is created using AI, labelling it or disclosing the fact that they have used AI.
At the moment, as you say, we have an influx of new developers adding their add-ons, and most are created by AI. Labelling them or having them disclose that AI was used, will definitely avoid downloading add-ons that could potentially break the forum or much worse.
I think that boat has sailed. Probably easier just to acknowledge the odd add-on that is still 100% human coded.
