@tobifant My read is that you are still framing things in terms of usefulness; where for many the usefulness of the output is not a determining factor, like @neil said. Put another way, you could have LLM code that is provably better (whatever that means) than the human equivalent and still not want to interact with it (on climate, stolen training data, vendor lock-in, or some other arguments).

"Is it useful" is not enough of a bar for many.

@Ozzelot @Neil Brown Any technology can be used responsibly and irresponsibly, and humankind has a track record of always starting out with the irresponsible one before finally setting for responsible uses, unfortunately.

I use local models, they take 150 Watts for inference, that's two conventional light bulbs. I reckon a lot of power was used to train them, but that's a one time effort. I consider this responsible use.

Book destruction - some tech bros been caught doing it. Of course that's evil but you can also scan books non-destructively.

Non-consensual use? Oh well, if you publish a book, you consent to anybody reading it.

Bottomline - for coding *assistance*, a small Qwen is all you need to get the job done father. Qwen already exists and you can run in with moderate hardware.

It will fail at just building an app by one click (but that's exactly the kind of apps we do not want on F-Droid or anywhere else), but it can help you get the job done faster while coding and still understanding what it is that you code.

See what I mean?

@nicolaottomano @nini I think many opposed to how LLMs are being framed and forced down our throats would agree they're a fitting example, though not based on the colloquial usage of the term but what they actually stood for:

The Luddites were members of a 19th-century movement of English textile workers who protested the usage of certain types of automated machinery owing to their concerns over worker pay, labour savings, child labour, working conditions, and output quality.

@tobifant
I disappoint… :(((

Still I'll dig into this.

Edit: AI Act art. 50 § 2 and guidelines read that “AI” providers and deployers shall mark any genAI content.

Now it gets murky if F-Droid or Debian are to be considered deployers¹.
IMHO IANAL opinion F-Droid would fall out of this definition (sans their client) as a store where you choose the apps to install. But Debian might not, as installing the distro you don't have the option to choose non-slop code.

1. AI Act art. 3 (4)

»

@neil

@tobifant
«
And on the same IMHO IANAL I would think that the Linux kernel falls right into the definition of “deployer”.

Of course this might hinge on the assumption that legal definition of “text” includes code (I'd argue yes).

@neil

> It is not for F-Droid to determine what F-Droid hosts

F-Droid already has an inclusion policy (f-droid.org/docs/Inclusion_Policy/).

I can't see the existence of such a policy changing (nor would I want to default to "anything goes").

So F-Droid already determines the boundaries of what it is wiling to host.

And those boundaries can - do - change over time.

Note that the repository system is open, in that developers can host their own repositories or use a third party repository, and users can choose what repositories to add.

> You can just choose not to use an app made by genAI?

This is hard to reconcile with an argument that people will not disclose their use of genAI. Both cannot be true.

In any case, developers who wish to use genAI could "just choose" not to submit to F-Droid's own repositories, and run their own, and make whatever decisions they like about the governance of that repo.