| Time |
Nick |
Message |
| 00:32 |
|
YuGiOhJCJ joined #luanti-dev |
| 00:56 |
|
chripo joined #luanti-dev |
| 01:04 |
|
YuGiOhJCJ joined #luanti-dev |
| 01:14 |
|
ndorik-zenith joined #luanti-dev |
| 01:34 |
|
sofar joined #luanti-dev |
| 02:29 |
|
SpaceManiac joined #luanti-dev |
| 02:51 |
|
ndorik-zenith joined #luanti-dev |
| 02:54 |
|
Warr1024 joined #luanti-dev |
| 03:01 |
|
_10KBps joined #luanti-dev |
| 03:34 |
|
ShadowNinja joined #luanti-dev |
| 03:35 |
|
ndorik-zenith joined #luanti-dev |
| 03:35 |
|
izzyb joined #luanti-dev |
| 03:37 |
|
rubenwardy joined #luanti-dev |
| 03:49 |
|
ndorik-zenith joined #luanti-dev |
| 04:00 |
|
rubenwardy joined #luanti-dev |
| 04:26 |
|
Alias joined #luanti-dev |
| 05:00 |
|
MTDiscord joined #luanti-dev |
| 05:13 |
|
mrcheese joined #luanti-dev |
| 05:48 |
|
ndorik-zenith joined #luanti-dev |
| 06:41 |
|
cheapie joined #luanti-dev |
| 07:16 |
|
dv^_^ joined #luanti-dev |
| 07:31 |
|
nore joined #luanti-dev |
| 08:49 |
|
ndorik-zenith joined #luanti-dev |
| 08:52 |
|
ndorik-zenith_ joined #luanti-dev |
| 09:05 |
|
ndorik-zenith joined #luanti-dev |
| 09:48 |
|
[MatrxMT] joined #luanti-dev |
| 09:59 |
|
Noisytoot joined #luanti-dev |
| 10:07 |
[MatrxMT] |
<Zughy> Given our AI policy, do we want to feature something like this? https://github.com/LibreSprite/LibreSprite/blob/master/AGENTS.md |
| 10:08 |
[MatrxMT] |
<Zughy> (I only use AI for spodadic audits so I genuinely don't know if it can be useful) |
| 10:09 |
[MatrxMT] |
<Zughy> *sporadic |
| 10:09 |
|
Lai0602 joined #luanti-dev |
| 10:34 |
|
Niklp joined #luanti-dev |
| 10:34 |
|
Juri joined #luanti-dev |
| 10:50 |
|
ndorik-zenith joined #luanti-dev |
| 12:49 |
|
ndorik-zenith joined #luanti-dev |
| 13:29 |
|
ndorik-zenith joined #luanti-dev |
| 13:32 |
|
vampirefrog joined #luanti-dev |
| 14:29 |
|
jonadab joined #luanti-dev |
| 14:51 |
|
ndorik-zenith joined #luanti-dev |
| 15:37 |
|
[MatrxMT] joined #luanti-dev |
| 15:50 |
|
Panquesito7 joined #luanti-dev |
| 16:45 |
|
SpaceManiac joined #luanti-dev |
| 16:46 |
|
erbium joined #luanti-dev |
| 16:50 |
|
ndorik-zenith joined #luanti-dev |
| 17:05 |
|
SFENCE joined #luanti-dev |
| 17:07 |
SFENCE |
sfan5: I usally inform here, that macos binaries were added. https://irc.luanti.org/luanti-dev/2026-11-01#i_6349281 |
| 17:07 |
SFENCE |
sfan5: should I mention you or somebody else next time? |
| 17:07 |
sfan5 |
no it's fine |
| 17:11 |
Krock |
waaaah merge conflicts are so annoyung |
| 17:51 |
|
klea joined #luanti-dev |
| 18:06 |
MTDiscord |
<siliconsniffer> Have you tried rebooting your computer yet? :P |
| 18:19 |
|
crazylad joined #luanti-dev |
| 18:57 |
|
MTDiscord1 joined #luanti-dev |
| 19:39 |
|
andriyndev joined #luanti-dev |
| 19:40 |
andriyndev |
https://github.com/luanti-org/luanti/blob/master/doc/developing/ai_policy.md. It’d be great to add the definition of “vibe coding”, because different people interpret it differently. |
| 19:42 |
sfan5 |
vibe coding is letting the AI write code without oversight or review, isn't it? |
| 19:45 |
Krock |
and a "work harder" prompt is probably not counted as oversight |
| 19:46 |
Krock |
(not implying that it happened here, just nonsense remark) |
| 19:47 |
celeron55 |
i stumbled upon this proposal for the grading of AI use. i think it's pretty good: https://www.bennjordan.com/blog/we-should-standardize-vibe-coding-ethics |
| 19:47 |
Krock |
defining rules is difficult and they may be enforced differently on a case basis |
| 19:51 |
Krock |
our policy seems to best fit into the "AI-dev-A" category ^ |
| 19:51 |
celeron55 |
in the case of the luanti core, out of those, the maximum allowable is probably AI-dev-A, and AI-UI, while considered lesser in that list, only applies as a mitigation in some specific cases |
| 19:51 |
MTDiscord |
<greenxenith> That article already fails in the first few lines by assuming there is no valid problems to be solved by the other generative content |
| 19:52 |
celeron55 |
feel free to write a better article. i'll then cite that instead going forward |
| 19:52 |
celeron55 |
it's close enough to reality in order to be useful |
| 19:53 |
MTDiscord |
<greenxenith> The AI-UI definition is also rather arbitrary, why is an AI-written interface important? |
| 19:53 |
Krock |
I assume this distinction was made because UI appears to be easier to test/verify (already by looking at it) |
| 19:53 |
celeron55 |
i think the main implication there is that in the case of AI-UI, all of the API and persistence related things were reviewed by an experienced developer. I'd redefine it as AI-ephemeral or something like that |
| 19:54 |
celeron55 |
or AI-frontend |
| 19:55 |
MTDiscord |
<paradust> "AI music solves the imaginary problem of there not being enough music on Spotify." Is so blind to humanity it makes me wonder if this was written by AI. It obviously solves the problem of people wanting to be able to express themselves with music, but not having the time or skill set to be able to compose it from scratch themselves. It doesn't have to be professional level or end up on spotify. |
| 19:56 |
celeron55 |
you can definitely grade AI assistance in music making very similarly |
| 19:56 |
celeron55 |
and anything else too |
| 19:56 |
MTDiscord |
<greenxenith> I certainly plan to write my own article about this at some point, but it is certainly not prerequisite to understanding that the article is not standing on good foundations. |
| 19:57 |
MTDiscord |
<greenxenith> Just because it fits your paradigm decently does not make it a good reference |
| 19:57 |
celeron55 |
it's basically "did you prompt it to do something specific (more points if you fed in some of your own material), and did you review the end result thoroughly" |
| 19:57 |
MTDiscord |
<greenxenith> Your own summary of the system is better than the system itself |
| 19:59 |
MTDiscord |
<paradust> I'm hoping the vibe situation with AI is temporary. Imagine if we had modern compilers but we still checked in the .asm they generated instead of the actual source, after manual review, because they occasionally generated code with subtle errors. It would be a nightmare. |
| 20:00 |
MTDiscord |
<paradust> Hopefully, what is currently vibe coding will become a higher-level description language. |
| 20:01 |
celeron55 |
i predict during the next decade people will start calling human-made things slop, as AI will be perfect on a level which simply doesn't invoke the feel of slop at all and humans are the ones being sloppy. which basically results in a world where everything polished is made by AI and if you want something made manually by a human, you have to look for something very "punk" |
| 20:02 |
ZenonSeth |
etsy but for devs |
| 20:03 |
celeron55 |
most people don't care about creator ethics on a level where they'd be angry of AI use if the result is better than otherwise. It's very similar as the current situation in software ethics - only the FOSS community cares about the details, everyone else just consumes regardless of how it was made |
| 20:04 |
celeron55 |
similar to* (AI wouldn't have made that mistake) |
| 20:04 |
ZenonSeth |
for the specific PR that andriyndev, who seems to have joined just to post that 1 message and left mentioned - that PR was 5k lines and growing. Even if it was reviewed by hand, that's irresponsible use of llms - pile on code and more code and more changes without feedback |
| 20:05 |
MTDiscord |
<greenxenith> Agreed, but that is not unique to LLMs. That is irresponsible development in general, and rules should not discriminate the tools used |
| 20:11 |
MTDiscord |
<luatic> I think it's an interesting idea to use LLMs for prototyping larger features, expecting that there's a high chance you'll just end up throwing the code away. I don't think that's really problematic per se. |
| 20:11 |
ZenonSeth |
personally I think subjective application of the rules is going to be the best people can do for a while, because to me LLM usage right now falls under sorites paradox - you can tell the really bad cases from the really good ones, but there's plenty inbetween that are subjective |
| 20:12 |
ZenonSeth |
luatic: For what purpose? did you spend a good half hour reading the code changes? is it worth it if they'll be thrown away? prototyping locally sure, but PR? |
| 20:13 |
MTDiscord |
<luatic> ZenonSeth: wasn't the PR marked as draft? |
| 20:15 |
ZenonSeth |
sure, yes, but it was still a public PR that was normally unreviewable, unless the details had previously been discussed a lot |
| 20:15 |
MTDiscord |
<luatic> Personally I don't really mind sloppy code (yes, even of the AI kind) in drafts, I'll just take a peek and if I don't find it interesting, spend my time elsewhere. Often I'll also just ignore drafts. |
| 20:15 |
[MatrxMT] |
<birdlover32767> for a website or something, ai-ui is an important distinction |
| 20:15 |
ZenonSeth |
eh fair |
| 20:16 |
ZenonSeth |
(in reply to luatic) |
| 20:16 |
[MatrxMT] |
<birdlover32767> for luanti, not so much (i guess the only real "ui" stuff is in the settings and main menu) |
| 20:16 |
[MatrxMT] |
<birdlover32767> my 2 cents |
| 20:16 |
[MatrxMT] |
<birdlover32767> (ish?) |
| 20:20 |
celeron55 |
i agree with luatic: it's perfectly ok to use an LLM to make a proof of concept to find out if something is possible, and also to post it as a draft PR for people to look at and see that if such a feature is desired, the cost is probably N lines touching the code the draft does touch. the disagreement here seems to be mostly whether there's any expectation of anyone reacting to it in any other way |
| 20:20 |
celeron55 |
than saying "that's a lot of lines, not worth it, i'll pass" or ignoring it entirely |
| 20:21 |
|
Panquesito7_ joined #luanti-dev |
| 20:23 |
ZenonSeth |
yea i think the trigger was that the pr was closed today with comment linking to ai-policy. And as i said that i got an email of more comments on the pr so it ain't over yet |
| 20:24 |
celeron55 |
i'd say responding "write it without AI and come back, and then i'll look at it" is not necessarily fair, because it is functional code and as such, anything a human does probably can't squeeze it into half the lines or half the dependencies or anything like that. asking someone to spend effort on something that won't be accepted anyway isn't fair, if they already provided the proof of concept which |
| 20:24 |
celeron55 |
already contains the important aspects |
| 20:26 |
celeron55 |
i'd say in general if someone posts a draft PR, the way forward from that is a concept review (either a full one, or one which weighs the scope vs. the roadmap), and if the concept is approved, then it is fair to ask for a version with significantly more oversight and optimization |
| 20:26 |
celeron55 |
if it's not approved, then it's best to be clear about it: no amount of manual coding will make it acceptable |
| 20:27 |
celeron55 |
(unless, of course, it's changed to an extent which completely alters the lines-of-code or dependency requirements, or whatever) |
| 20:29 |
celeron55 |
i'm not sure if i'm making my point understandable at all :D i'm trying to say there's value in a proof of concept, regardless of how it's made, and it's wise to utilize that value, and to utilize that value doesn't mean merging the PR |
| 21:37 |
|
panwolfram joined #luanti-dev |
| 21:44 |
|
buddyholly joined #luanti-dev |
| 22:20 |
[MatrxMT] |
<Zughy> I don't understand how the new version is cleaner than the old one |
| 22:20 |
[MatrxMT] |
<Zughy> https://i.imgur.com/ofWtGJF.png |
| 22:20 |
[MatrxMT] |
<Zughy> it feels very redundant |
| 22:27 |
|
panwolfram joined #luanti-dev |
| 22:36 |
|
ndorik-zenith joined #luanti-dev |
| 22:41 |
rubenwardy |
ah yes, "Content not available in your region" |
| 22:42 |
[MatrxMT] |
<Zughy> I'd link GitLab but it keeps me out because it says that I've reached the rate limit. By opening 5 pages and not using webhooks, AI or anything :))) |
| 22:48 |
[MatrxMT] |
<Zughy> RE AI: it seems to me that nobody would look at a 5k PoC anyway, no matter if it's AI or not, so I can't see the problem (nor the reason to discuss it further) |
| 22:50 |
[MatrxMT] |
<Zughy> Also, call me romantic of whatever, but I think we should build community, helping people getting better at their skills by providing insightful reviews. It's an investment for the future. On the contrary, |
| 22:51 |
[MatrxMT] |
<Zughy> what's the purpose of reviewing code made by a machine? I mean, there's no investment, no community building |
| 22:52 |
MTDiscord |
<greenxenith> If an active member of the community finds something they want to fix, but has no coding skills, I see no reason to require them to learn programming just to fix it. They are invested, and are part of the community. The purpose is to ensure someone knowledgeable has verified it works and looks ok. |
| 22:52 |
[MatrxMT] |
<Zughy> Let alone all the ethical issues that we're ignoring right away - something I find absurd |
| 22:53 |
[MatrxMT] |
<Zughy> That is slop, green. No coding skills, just prompted and fingers crossed |
| 22:53 |
[MatrxMT] |
<Zughy> so more work for maintainers |
| 22:54 |
MTDiscord |
<greenxenith> How much do they need to know for it to not be slop? |
| 22:55 |
rubenwardy |
~~it's less about the knowledge and more about the expensive student loans~~ |
| 22:59 |
MTDiscord |
<greenxenith> Im playing the extreme end of the scale here to say the rules dont adequately account for all cases. I dont think blind vibecoding is helpful most of the time. I think distinguishing the tool is incorrect. "Can you take responsibility for this code or not?" |
| 23:05 |
|
ndorik-zenith joined #luanti-dev |
| 23:09 |
|
[MatrxMT] joined #luanti-dev |
| 23:09 |
[MatrxMT] |
<Zughy> I think that that is a slippery slope |
| 23:11 |
MTDiscord |
<greenxenith> And I think it is a slippery slope to ban AI or demand disclosure for any use of it |
| 23:11 |
MTDiscord |
<greenxenith> I should really write about this, I tire of explaining this frequently |
| 23:15 |
MTDiscord |
<wsor4035> are we taking bets on that? |
| 23:16 |
MTDiscord |
<luatic> Zughy: re the screenshot you've shared above, looks reasonable, depending on what you want. if that code is supposed to hide a nametag by making it transparent - and nothing else - then what the change does (preserve current RGB) makes sense. |
| 23:17 |
MTDiscord |
<luatic> it fixes a minor bug where the color would otherwise be lost (the C++ reading code always replaces the entire color, not just a single component). |
| 23:18 |
MTDiscord |
<luatic> github is getting on my nerves with a banner pestering me to add payment info, even though i only use their free stuff 😒 |
| 23:18 |
MTDiscord |
<wsor4035> got a picture? i think all the banners ive ever seen ive closed and they went away |
| 23:32 |
[MatrxMT] |
<Zughy> luatic: yes but now I have explicitly state that r, g and b are equal to what they used to be |
| 23:32 |
[MatrxMT] |
<Zughy> *have to |
| 23:33 |
[MatrxMT] |
<Zughy> I understand it for e.g. HUD's alignment, but colour of nametag, c'mon.. |
| 23:33 |
|
panwolfram joined #luanti-dev |
| 23:34 |
[MatrxMT] |
<Zughy> I don't expect the game to crash if I only declare the sun table instead od stars, moon etc. in set_sky(). Same logic |
| 23:34 |
[MatrxMT] |
<Zughy> *or |
| 23:43 |
MTDiscord |
<luatic> Zughy: ah, you're complaining about the engine warning? i think that's useful actually. |
| 23:44 |
MTDiscord |
<luatic> because you're replacing the color wholesale, and if you're forced to specify the entire color that becomes obvious. (under the hood, the engine is reading a "colorspec".) |
| 23:44 |
MTDiscord |
<luatic> it's not "incremental" in the way set_properties() is. |
| 23:44 |
MTDiscord |
<luatic> what your previous code was doing was equivalent not to setting just a = 0, but rather to setting the entire color to {r = 0, g = 0, b = 0, a = 0}. |