https://tintotint.eu/whacky-corner/f-droid_slop/

[shtrom@piaille.fr]

@fdroidorg It'd be interesting to see how long apps go without maintenance vs. whether their level of LLM involvement.

[hipgnose@mastodon.social]

@fdroidorg I fully agree with the blog post!

I've always liked to check out the new apps and read the "This Week in F-Droid" new app list, but it seems that nowadays it's A LOT of low effort, weird apps. There is no joy in that.

[cryptgoat@fedifreu.de]

@fdroidorg Pretty much confirms what you could tell already by looking at many recent new apps. Checked many of my apps I have been using for years now and the results certainly raise an eyebrow.

#FDroid is all about #FreeSoftware and transparency, yet it is hard or or impossible for non-techies to check if they download an #AIslop app from #FDroid or something entirely human made. I'd still like to see some kind of #AI / #LLM disclosure for apps being submitted to the F-Droid repo but it's hard to find a good approach.

Main problem is that many people just lie about their AI usage or change their approach after a successful submission. Happens all the the time on the /selfhosted subreddit.

Btu I see this is already being discussed: https://gitlab.com/fdroid/admin/-/work_items/641

Edit: Added correct issue.

[frisk@woof.tech]

@fdroidorg While 102 apps is certainly not representative of entire F-Droid library, I still think the report is fairly indicative of direction we are going in/current situation in FOSS.

Mostly because while it could be possible that in that date surprising amount of creatures using AI to build their software decided to suddenly push their app to F-Droid, I find it unlikely that the time of pushing an app itself would play a major role in adding bias to general result.

And I can only imagine checking each cycle would be very time consuming.

[fdroidorg@floss.social]

@cryptgoat The note on top should have guided you to the proper issue

[kayohtie@blimps.xyz]

@fdroidorg I'm put off the article by some updates saying a project is "mostly AI"... That has existed since before it was even open sourced in 2012.

That said I haven't used SkyMap in years, so I don't know how much it's changed. Could be recent fixes were LLM, which is upsetting if true, but the classification being "mostly AI" instead of "mostly human" comes off as more "this is just my 30 second vibes off a thing rather than looking into the history" in a way that makes me uncomfortable with any other "mostly AI, cause it looks like it could be" classification that person makes _without_ the visible proof of such. They even say it's hard to tell off text alone but proceed anyway, like... I dunno, feels antithetical to its own thesis?

So many projects using it just suuuucks though.

The standout for me has been the enormous influx of "material you music streaming apps" that half the time don't even say they're just ripping music off YouTube Music.

[b4ux1t3@hachyderm.io]

@fdroidorg it’s funny because the writer said it themselves. They are a junior developer. Of course it’s hard for them to tell slop from not. They’re right that there are signs.

But weirdly, “the code is wrong” isn’t one of those signs. It’s kind of the opposite.

The code is “weirdly right, but missing things that are obvious to most human developers”.

Like, it’ll have accessibility features bolted in, but not in the way humans bolt on accessibility; it’ll have a bunch of massive reimplementations of things like screen reader support.

It’ll have tests. Lots of tests! And about half of them will be like “hey let’s test the same thing we just tested but with the input changed by one character, despite that having no difference to the function”. Manual fuzz testing in the test suite.

The author is right that there are signs and they’re not perfect indicators; the AI learned these behaviors from human code, after all.

But man, it’s not…that hard. I know _instantly_ when one of my coworkers just told Claude to read the ticket and do it. The thing that’s harder to tell is not whether something was vibe coded, but rather whether it was reviewed by a human.

You can make “good code” with LLMs. They’re a tool. An expensive, wasteful tool, which is hard to get exactly the right output from, in the same way that trying to paint a painting with the mechanical arm of an etch-a-sketch is hard.

But they won’t get there on their own, and I’d wager most relatively experienced developers can easily spot the difference between “this person prompted Claude and walked away” and “this person prompted Claude and then iterated”

[cryptgoat@fedifreu.de]

@fdroidorg
Indeed. https://gitlab.com/fdroid/admin/-/work_items/641

[simx72@tkz.one]

@fdroidorg I think it would be a good idea to create an online "comittee" of volunteer people to tag the apps. Or create an Fdroid login site and have fdroid accounts just to submit stars and comments (like play store)?

Maybe this exists already. If you know it, please tell me ^^

[felix@wandering.shop]

@fdroidorg That write-up is so verbose, I gave up after a couple of subheadings or so. It wouldn't surprise me to learn it's extruded. But the problem is real. Once again: please make vibecoded apps an anti-feature.

[simx72@tkz.one]

@Frisk @fdroidorg I think that is something Fdroid users could do.

I know democracy is not always the best solution, but in this case, it could be very helpful

(my other answer)
https://tkz.one/@Simx72/117275283314435522

[xz@ieji.de]

@fdroidorg unmaintained apps are a bigger problem

[nube@verse.averagedood.xyz]

@fdroidorg@floss.social IzzyOnDroid has a firm and clear policy, while you who must lead, have a weak and ambiguous posture...

Are you afraid of receiving a negative response from the vibe coders and losing the number of apps available? tbh, does it matter? F-Droid was good before them, it can be good without them