Hi,
I’m a former vim user looking for a new text editor due to the recent use of Claude as both a generator and reviewer of code added to the codebase.
Does anyone know whether kakoune as a project has, or will have, any statement on accepting LLM generated contributions to the project?
I’m not going to try to convince anyone to disallow LLM generated code in their codebase, I’m just trying to avoid surprises down the road, given that my text editor is a pretty central part of my workflow.
What exactly is your issue with AI-generated contribs?
Since you asked,
In short, I want as few of them as possible in software I use.
I do not believe any LLM model is up to the task of producing good software. Snippets that are useable? Possibly.
Beyond that there are copyright issues that haven’t yet played out, the unlicense is great, but it still is dependent on the person who created the code having the right to dedicate their code to the public domain.
I’ll acknowledge that appropriately vetted, edited, and tested code created with these tools in the loop is probably not something one can “prove” was used enough to enforce a “full ban” of any contributors having any level of use of them.
I’m more concerned about the mess that happens when these models are used for both generation and review of things. There’s increasing evidence that ongoing use of these tools creates an undermining in the confidence of the user in their own knowledge. Trusting these tools to “check their own work” is, as far as I can see, a relinquishing of the stewardship of a project over to tools that are fundamentally unsuited to the task.
That said, this isn’t my project. I fully recognize that the project leads are well within their rights to allow anything they’d like. I’m not even a long-time user with any kind of investment in the project.
I’m not here to shame anyone for their choices. I’m trying to rebuild my personal toolchain, and as any major tool shift is bound to create headaches for actually getting work done until one has settled into the new tool, I am hoping to find out whether this one is one I’m likely to have to replace down the road or not. Because frankly, in trying it out for a bit, it feels like I’ve upgraded. Once I got my head around it, it feels like an improvement on vim. So I’d like to invest the time in building my replacement toolchain around it.
I’m simply trying to make an informed choice, and as I can’t find anywhere where a statement as been made, I figured I’d ask here.
Out of curiosity, what are your reasons for trying Kakoune instead of any other alternative to Vim/Neovim?
Upon realizing Kakoune is doing this I have fixed my version and am migrating away from using it, which makes me really sad.
I find the question “What exactly is your issue with AI-generated contribs?” rather horrifying. Do you not see the slippery slope that comes with excessive convenience? Look not at what it accomplishes in changing your codebase; instead look at what it accomplishes in changing your perceptions/habits/acceptability. Each action you take is a form of programming your own internal software as a human being as to what you find normal/acceptable in this life we encounter.
The use of these kinds of things is deferment of one’s own understanding (strength as an individual) to some complex piece of software you did not write and will never be able understand/vet. The intended purpose of it’s use is to answer a question/proofread something you are too lazy to figure out/do yourself. By design of removing the effort involved in attaining that knowledge/accomplishing that task, this approach inheritly makes people dumber the more they utilize it. That’s not an insult toward anyone using such tools, it is a plea of acknowledgement.
How many people do you know whom can NOT reliably drive around the city they have lived in their whole lives without use of a GPS? I know many…
How many programmers actually know how binary computation happens and not just how to compose c_like(functions(together()))? I have met shockingly few…
If anyone is using LLMs to autogenerate the code to make their software, I have no idea what they are even doing anymore/what they care about. To me all that act signifies is ‘giving up’ and when someone makes ‘giving up’ acceptable, I have little hope that there is any potential for growth/improvement in whatever that thing is (or whatever that person does with that mentality). Even in terms of documentation, if someone cares so little about writing the documentation to communicate to me what their software does and how to use it, how much care/effort went into communicating the proper instructions to the machine? (probably as much as went into the documentation)
But forget software, having such little care towards any discipline a person embarks upon in life just really, really sad. If one degrades the integrity of their own passion they are doing a disservice not only to themselves, but to the beauty created by the communities whom enjoy such passions.
Have you ever tasted a delicious meal homemade from scratch by an amazing cook who put a bunch of love into every part of making it?
Have you ever eaten fast-food?
Both accomplish the same immediate need. One provides nurishment/life while the other provides you something else entirely. Doesn’t matter if you are a few months old or 50 years old, over time the act of consuming one or the other will induce their correlatory effects all the same.
You say “Upon realizing Kakoune is doing this”. Is there any evidence that Kakoune is accepting “AI” generated contributions? As of right now, I have a lack of any kind of answer as to a policy either allowing or disallowing them.
While for the most part I believe we are in agreement about the dangers of AI use are, You may also be inferring tone beyond what strlen actually asked here. Asking me for clarification on what I have against AI generated contributions is not, by nature, an argument for allowing them.
Well, one influence was coming across skrewtape’s piece describing the editor as “punk rock” Kakoune, a Punk Rock Text Editor - Screwtape's Notepad
Secondly, when I asked fedi for recommendations for alternatives for the combo of vim and vimwiki that runs much of my life, multiple people said that they’d set up wiki systems of one kind or another along with their Kakoune setup.
So, it’s one of the first things I’ve tried (after poking at emacs+orgmode and it still not working with my brain). I’ve always been someone who was “overly reliant” on visual select when using vim, and frankly the way selections work on Kakoune is awesome.
dude, calm down, just asking out of curiosity, rofl. I wanted to understand whether the reasons are ideological or pragmatic… Personally, I have my reservations about using AI for software dev because I’m a cybersec professional and I’ve seen some risks involved in…
Huh?
I was in no way upset/angry, nor was I inferring anyone’s viewpoint other than my own. My question of being able to see the danger of convenience was a genuine one. My explanatory response was not describing what I think your opinions are (I do not know anything about you and cannot do that), but to introduce my ideas to any reader who may come upon this thread in the future to observe/consider/reply to, as that’s how forums work.
There was not aggression in that message, there was however conviction and passion.
I was sharing my viewpoints on the topic because this is the only thread in all of these forums to critique LLM usage at all, and I wanted to contribute to that discussion because I find the overwhelming lack of critique and acceptance of these things both shocking and dangerously foolish. That is why I come to discussion forums, to discuss topics, but it seems that topic does not actually want to be discussed/explored.
https://discuss.kakoune.com/t/ai-prompt-stuffing-for-the-kakoune-asciidoc-documentation
@socialgaff That thread and the general acceptance/movement towards these kinds of things on the forum indicate to me the general direction of the software. It’s more a feeling than an official declaration I am going by, but my gut always seems to be much smarter than me.
This particular topic is really, really difficult to discuss in a useful, good-faith way, because there’s so much bad-faith discussion of it going around. Not just from people with a direct financial interest, but also because over the last twenty years people have been taught that the most effective way to defend against something undesirable is to filibuster it. And so, any discussion of the thing in question immediately attracts pitchfork brigades from both directions.
What I will say is this: the strength of LLMs seems to be in making things mediocre - if you have something crappy, it can boost it up to mediocre levels quite quickly. In a world where so much effort is spent on making small adjustments to compensate for small adjustments made elsewhere, I can see the value in that kind of automation - it doesn’t matter much if it’s mediocre, it just has to keep working until the next round of small adjustments is made.
From that perspective, Kakoune does not seem like fertile ground for LLM usage: mawww shows very good taste in the changes he accepts, or makes himself, and because it’s not a commercial endeavour, there’s no temptation to get new features done to juice the quarterly earnings report or whatever. Meanwhile, there’s no benefit to having an LLM scaffold the boilerplate for a plugin, because there is no boilerplate. The Kakoune ecosystem is built from parts that are either ultra-polished or ultra-scrappy, there’s not really room for mediocrity.
So I’m not really worried about some LLM-assisted commits winding up in Kakoune’s repo, because I trust mawww to only accept things that make sense. I am worried about the Kakoune community maintaining its culture, so I take time to help people who are polite and interested, and I ignore people who seem to care more about the destination than the journey — regardless of whether they use LLMs or not.