Why should they ? You are entitled to know only what they want you to know.
You can choose to ignore them if you want.
If they choose not to, and post LLM generated garbage (even if their intent was a translation), then they can get treated like everyone else who posts LLM spam…
EDIT: Though pulkomandys solution is losless, more elegant, and much more usefull. So let’s just do that.
1 Like
Both Google Translate and DeepL use LLMs these days. I regularly use DeepL to check grammar and idiomatic expressions, sometimes to translate a sentence or find a better word choice. I hope the Haiku community won’t dictate to its members which tools they are not allowed to use.
4 Likes
Agree strongly. We tend to be quite fearful of Linux-style fragmentation, but a larger selection of repositories and websites will not necessarily lead to separate distros. Not everything needs to be on Haikuports.
I honestly didn’t wanna weigh in on this sh*tshow but I am shocked and deeply offended by the utter contempt shown in the original post at people porting apps.
No email client that actually works.
We have 4!
No video calling app.
Telegram
No office suite
LibreOffice, Calligra (and whatever else is in the KDE Office Suite), GoBe (on 32bit), AbiWord
No Mastodon client
Tokodon
no maps, no ebook reader
Both false claims
Too bad that in twenty years this approach has produced, roughly speaking, three ports of already existing software and a GTK theme.
Thank you, I alone ported at least 2 dozen apps on my own and updated/maintained/fixed many others without asking for anything back and this is the thanks I receive???
I totally feel like porting more things now, thanks for the encouragement. /s
Why haven’t you deleted this blatant inflammatory bs from your original post yet?
Let me guess, you want to save water and energy that would be lost from a post edit. Riiight.
6 Likes
I’m probably on the extreme anti-LLM side, if that weren’t obvious. 
My first preference would be for someone on the pro-LLM side to start a new Haiku Depot-compatible repo that proudly publishes LLM-generated apps, so that haikuports could put up a “no generated code”policy. That would be ideal.
My second choice would be for haikuports to institute labeling, maybe a multi-level version as I mentioned above, with an associated filter option in the Haiku Depot app.
3 Likes
I joined in the last two weeks, found a bug, and fixed it with help from long-term contributors.
I saw no barriers. In fact, I found it an amazingly helpful community and I can’t wait to do more.
7 Likes
This post seems to imply that anyone who uses a LLM agent is aitomatically a vibe coder and there only are fully LLM generated applications and fully manually typed applications. Obviously that’s not the case.
What would you do if GCC contributors started using an AI assistant? Would you remove GCC from haikuports? Firefox? LibreOffice?
1 Like
I would genuinely be in favor of sticking to old versions of gcc for the base system if that happened.
4 Likes
Actually, the list of possible categories I’d posted above dealt with that explicitly; there’s definitely a difference between occasional / regular LLM use and full-on vibe-architecting / coding.
From my perspective, I don’t want to use any of it but some folks may be open to different levels of use so a variable in the port recipe file with multiple possible values seems the best solution to me.
Personally, I would drop gcc over that; I stopped updating Firefox on the release before they started going AI. And yeah, if haikuports were my project I’d freeze the gcc version to the one prior if they did that.
So we’re probably all fortunate that I’m not in charge of haikuports.
Yeah and then maintain it by yourself? Good luck with that!
I’ve spent extensive time working on the guts of a commercial OS that uses gcc4 to build. It’s really not that much of a problem, especially since - per my understanding - Haiku avoids language features that aren’t buildable with gcc2 anyway. (I could be wrong about that.)
So using an AI agent or coding assistant automatically makes the existing GCC developers a brunch of incompetents?
If I used and AI assistant for my numerous haiku code contributions over the years, would that make those contributions less valid?
Anyway, using outdated software isn’t a very wise choice. Especially a web browser.
1 Like
No, it does not make them incompetent but since their contributions are no longer actually their contributions, it does make those contributions untrustworthy.
The trustworthiness of the output is my main concern; I know for others here the main concern is intellectual parasitism or climate impact. In all three of those cases, you can take a stance anywhere between one-drop-is-too-much and full-on-YOLO and make a case to defend it.
This is why I think flagging the level of LLM-generated content is probably the best choice; it lets the user follow their own personal preference.
3 Likes
Use of LLM in a software should not be an issue, given the developers are competent and the code quality remains the same. It’s just a tool after all.
Obviously, a vibe-coded app with a single prompt and another app that uses highly specialised prompts at each step and audit will not be the same.
2 Likes
You know who has been maintaining a fork of gcc 2.95.3 all those years?
I can’t say I will enjoy the extra workload. But if it’s needed, I’m willing to go there, and since I clearly am not alone, I think it will be OK.
No, but that would make them and their decisions dangerous for software architecture, security, stability, and long term maintenance. At least that’s what I get from the studies I read about LLM and the feedback of various people I trust about this. LLMs generate “OK” code, the kind you’d expect from a junior developer maybe. But whenever you ask for something a bit complicated, that no one has done before, it fails miserably. When the code looks great, it means you didn’t look close enough.
Sure, if you work alone on a project, this could be helpful (but you have to ignore all the previously mentionned issues with energy usage, the DDoS of many websites on the internet by the training data scrappers, the underpaid labor of people tagging data and preparing it. Which is quite a lot to ignore for me). Anyway, let’s say you’re OK with that, it can be helpful if you work alone, to fill in the parts that you don’t master yourself. But, in Haiku, I have often found that other people would know better than me about different areas of the code, and we could work together to fix, essentially anything. So, I see no reason to replace this social interaction and learning experience with a machine. Maybe I like too much to write the code myself, and spend years finetuning things. Probably not everyone is like that. We are in a society where everything must always go faster and faster. I’m not interested. I enjoy the journey more than the destination.
8 Likes
I too (mostly) enjoy writing code, but that doesn’t mean AI assistants aren’t useful, even if just for quickly exploring various solutions.
What you wrote makes me think you’ve never even tried using a LLM / AI assistant, and only speak by what you heard around.
Reading studies and listening ro people is great, but trying stuff first hand Is another thing.
There is no need to try [drugs] or AI before dismissing it as not usefull to you.
Sure, some people say they feel faster with it, but then again, the science sais they are wrong. And additionally, even if it was true, development speed is literally not what we optimize for. If you want to use llms, knowing the risks, and the impact, If you still want to do it despite all of that… Then just do it somewhere else. Don’t do it in Haiku.
I don’t understand what’s so difficult about that. Especially if, as some people here claim, it is used “everyhwhere”, then just do it somewhere else. It’s just not wanted here.
5 Likes
Not all Haiku developpers agree on that. Jackburton clearly doesn’t. I know Stippi uses or used LLMs as well. They both have commit access and a voice as much as you and me in this discussion, before deciding “it’s not wanted here” globally for Haiku.
And there’s no rule that the currently most active developers have more say in these things than less active ones.
2 Likes
Are the existing code acceptance guidelines wrong then? Has the copyright issue been miracilously solved?
1 Like