IRC logs for #openttd on OFTC at 2025-08-27
⏴ go to previous day
01:54:51 *** ChanServ sets mode: +v tokai
02:01:50 *** tokai|noir has quit IRC (Ping timeout: 480 seconds)
02:03:22 *** herms has quit IRC (Quit: bye)
02:13:25 *** raxitocl has quit IRC (Quit: Page closed)
02:45:22 *** Wormnest has quit IRC (Quit: Leaving)
03:16:02 *** Zathras_11 has joined #openttd
03:18:56 *** Zathras has quit IRC (Ping timeout: 480 seconds)
04:39:18 <DorpsGek> - Update: Translations from eints (by translators)
04:42:21 <yiffgirl> i was going to post something more incendiary but the new proposal seems fine
05:22:01 *** dh1 has quit IRC (Quit: My Mac has gone to sleep. ZZZzzz…)
05:29:43 <yiffgirl> the four horsemen: calling things that don't use llms "ai", claiming things that could be done since the 90s are only possible with llms, deskilling, and psychosis
05:30:26 *** dh1 has quit IRC (Ping timeout: 480 seconds)
05:35:00 <andythenorth> I think you inverted the first one 😛
05:38:54 <kuhnovic> Funny how some people can only see the bad side of LLMs. Makes you wonder if they ever used them at all.
05:53:23 <andythenorth> the specific, precise, issue here is GPL
05:56:59 <andythenorth> without case law it's unclear what the copyright treatment is of LLM-generated code per jurisdiction
05:57:18 <andythenorth> and without copyright, we can't accept LLM-generated contributions
05:59:21 <andythenorth> I use GPT generated code in Iron Horse becuase (1) UK case law sets a precedent that if I caused the output to happen, I'm the author, so I'm prepared to gamble on that being GPL compliant (2) GPT ToS indemnify me from upstream copyright holders brining any case against me as author
06:01:12 <andythenorth> OpenTTD can't operate on that basis AIUI
06:27:26 *** dh1 has quit IRC (Ping timeout: 480 seconds)
06:44:16 *** dh1 has quit IRC (Ping timeout: 480 seconds)
06:58:19 <peter1138> Correct, I have never used them, because I do not want to use something that is based on the scraping other people's work without consent. It's not just the licence issue, it's an ethics issue.
07:00:10 <peter1138> And there is the whole thing about it being shoved in our fucking faces everywhere.
07:10:12 <peter1138> You migh want to read that indemnity clause, because it's very likely that actually indemnity works that other way, and you CANNOT hold them liable.
07:12:22 <peter1138> I cannot see a world where a company would willing indemnify all of its users, never mind that even if it did what does that really mean.
07:18:44 <andythenorth> oh yes, unsurprisingly, I am liable
07:18:53 <andythenorth> either they've changed it, or I read it wrong last time
07:19:18 <andythenorth> the current Anthropic suit means that both options are quite plausible
07:41:01 <peter1138> Refurbished Mac Mini for development testing?
07:41:18 <andythenorth> got an M1 macbook pro here that needs sold 😛
07:41:47 <andythenorth> it has a cosmetic hole in the display bezel though, which limits ready buyers 😐
08:07:48 <ahyangyi> yiffgirl: That proves you havem't even tried, and are talking about things you don't even know.
08:08:27 <ahyangyi> For or against AI usage is one thing, spreading misinformation is another
08:08:34 <ahyangyi> You are making false claims here.
08:11:05 <ahyangyi> Go to read some books to learn about how computer scientists in the 1960s actually thought about Artificial Intelligence, instead of relying on sci-fis, please.
08:11:45 <andythenorth> I don't know why LLMs get some weird treatment
08:11:51 <andythenorth> they're just an extension of prior art
08:12:04 <LordAro> ahyangyi: are you saying there aren't a lot of things that LLMs are being used for that are otherwise not possible?
08:12:21 <LordAro> wait, triple negative
08:12:49 <LordAro> ahyangyi: are you saying that LLMs are not being used for things that have been perfectly possible since the 90s?
08:12:55 <LordAro> because they definitely are
08:12:59 <andythenorth> we built a semantic tokeniser and vector hyperspace on the cheap about 15 years ago, it just didn't have the recursive capability, nor did we figure out we should connect the prompt to the same analysis
08:13:11 <andythenorth> not wildly different in concept to a modern LLM
08:13:30 <andythenorth> our capital budget was 2 cheap dell desktops and hadoop
08:13:36 <andythenorth> so our training ingest was slimited
08:15:10 <ahyangyi> LordAro: It was a concrete example I posted in the Github issue: "Cursor's ability to move your cursor to a different line it thinks you want to edit next"
08:15:10 <ahyangyi> Which, I claim, is not possible without using a kind of statistical tools.
08:15:10 <ahyangyi> And, while older statistical methods could in theory produce similar effects, they did not actually exist.
08:15:51 <peter1138> Why is it meant to be some kind of gotcha that people against using this stuff haven't used it...?
08:16:17 <ahyangyi> No, i'm not against the "against AI" stance
08:16:26 <ahyangyi> I'm against the misinformation they are spreading.
08:16:33 <ahyangyi> You can be against something without spreading misinformation.
08:18:36 <LordAro> none of the 4 things yiffgirl listed are outright wrong, afaict. embellished, perhaps, but not wrong
08:18:59 <LordAro> as such, i object to "you are making false claims"
08:19:13 <andythenorth> what is ground truth anyway?
08:20:20 <ahyangyi> LordAro: If you ignore the fact they were commenting on my comment and not in general.
08:21:16 <ahyangyi> And I find it very offensive that, after I already declared my support for the "against AI" policy, to receive a "four horsemen" comment on me
08:21:18 <LordAro> i know nothing of this thread bullshit
08:23:27 <andythenorth> not sure it was about you
08:23:30 <ahyangyi> And still, no matter what you think, listing "calling things that don't use llms 'ai'" as one of the horsemen is outright wrong
08:23:36 <andythenorth> think it's more just a noisy chat
08:24:10 <ahyangyi> AI research happened about 80 years before the first LLM
08:25:17 <peter1138> It's easy to reply to the wrong thing, or deliberately reply to something vaguely relevant to link it. I doubt she was actually having a go at you personally.
08:25:49 <LordAro> ah, a common or garden internet chat misunderstanding
08:25:58 <LordAro> not an LLM misunderstanding ;)
08:26:15 <dh1> also we say horse PEOPLE now
08:26:15 <andythenorth> meanwhile in my LLM
08:26:23 <andythenorth> been waiting for 5 minutes
08:27:32 <dh1> i think i'm against llms in the same way i'm against fifa and nestle and capitalism. i.e. i recognise it is broken and flawed and extractive and awful but also recognise that only system change can be effective at addressing all of that, so individual boycotts feel utterly pointless and meaningless and yelling into the void. which is a nice comfortable copout position
08:27:50 <dh1> but also i have not contributed to openttd in any way in 25 years beyond sitting in here or on forums and occasionally whining about a button
08:28:59 *** toktik has quit IRC (Quit: Ping timeout (120 seconds))
08:31:09 *** toktik has quit IRC (Remote host closed the connection)
08:34:44 <dh1> from an ethical perspective it seems pretty reasonable and right to draw a line. i dunno if the quality perspective holds up though - if there is already good governance about what goes into main (i.e. n human people reviewing each line of code) then it's kinda irrelevant where it came from
08:35:44 <dh1> it'd be mildly interesting to see openttd fork into llm vs non-llm development
08:36:42 <dh1> i guess from a slop perspective it's a volume thing. it's far easier to create lots and lots of slop than it is to review and reject it
08:39:43 *** fairyflossy has joined #openttd
08:39:43 <fairyflossy> My question with banning AI-produced code is also how do you tell
08:39:55 <fairyflossy> Like sure if there's obvious weirdness in it it's one thing
08:40:03 <dh1> yeah, it feels like it is mostly a voluntary compliance kinda deal
08:40:05 <fairyflossy> And having an official policy on it is good
08:40:17 <dh1> but that is still normative and important i think
08:41:06 <dh1> like there is some sort of social process and accountability for all contributions already and this is just an extension of that?
08:41:29 <fairyflossy> It's good to make a decision on it, and say "this is our official stance on the matter", but I don't believe it's possible to actually completely ban it. Someone with enough dedication will be able to slip through code made with AI. Perhaps to the point of where they've actually done a lot of work with it themselves, but I feel like it's also important to consider that
08:42:25 <LordAro> fairyflossy: That'd be very useful to define in some fashion, IMO. Mostly I think it's based on the PR description itself. The code itself is generally... code. People write all sorts of nonsense. Sometimes there are tells (usually, the quantity of comments), but they're by no means conclusive
08:43:18 <ahyangyi> I think the practical usage is that developers can outright reject obvious AI creations and save time and energy.
08:43:34 <LordAro> i.e. The original PR that set all of this off probably would've gone unnoticed if the description hadn't been AI generated - especially given the author says the vast majority of the content was hand written
08:44:27 <fairyflossy> In practice, any stance on AI is going to be "Hey, we don't want you using AI, this is why we don't want you AI, please don't use AI, if we suspect you use AI we will deny you."
08:44:27 <fairyflossy> And then people will either respect it and not use AI, or not respect it and make an effort to mask that AI was used.
08:45:19 <fairyflossy> Which I'm not saying that defining a stance on AI is bad, I agree that it's a good idea, but it's not going to be a magic bullet against AI. It just means it'll be harder to tell what's AI
08:45:26 <LordAro> peter1138: I have no particular reason to doubt them. I've never known an LLM generate FIXME comments, certainly
08:46:08 <dh1> i always find it interesting (with no judgment) what people choose to care about in here. like this is a game based substantially on someone else's IP and where a bus is the same size as a stadium
08:46:46 <fairyflossy> My personal stance on AI code is that I do not like it. Both from an ethical perspective on how it learns, a concern on the impact it has on the environment and economy, but also the impact it has on one's understanding of what they're doing and what one learns
08:47:50 <fairyflossy> Generating AI code is easy, but you don't really learn from it. You don't trial and error or figure stuff out yourself. The more you make with AI, the more you grow dependent on AI. You won't be able to really debug it or see what's going wrong or miss obvious problems because you didn't create these problems yourself
08:48:03 <dh1> the what one learns point always confuses me because i have learned a ridiculous amount from LLM stuff - mostly from debugging LLM code
08:49:02 <dh1> it won't be long before people start getting candidates to review and pick apart LLM code in interviews
08:49:53 <LordAro> it's not especially different from "find all the bugs in this piece of code", which is definitely already being done
08:50:05 <fairyflossy> It depends on how complicated the code is, I suppose. If you generate code to do a thing, you're not looking up how to do this thing yourself. You're not going "Okay, I need to do this, how do I do this", you're going "Hey, I want X result" and only if the result isn't what you want do you dive deeper - or just tell the AI "no, no, like this, not like that"
08:50:37 <dh1> the worst thing about LLMs is it has actually DISRUPTED technology stuff and that pisses me off because i am often around people who unironically use the word disrupt
08:50:43 <dh1> and i want them to be wrong about all things ever
08:51:00 <dh1> fairyflossy sure but that is also true about my car
08:51:13 <fairyflossy> Yeah, but are you a car builder?
08:51:23 <LordAro> dh1: don't get me started on automotive software
08:52:07 <dh1> i don't think those are analogous analogies
08:52:18 <dh1> i think you're a computer driver
08:52:48 <ahyangyi> A computer driver, a hard disk?
08:53:16 <LordAro> computer drivers are just something that's gone wrong when the wifi doesn't work, right?
08:54:16 <dh1> you know how some cars pipe in engine sounds to make people feel like theyre driving a cooler car than they are
08:54:20 <dh1> i want that but with dialup noise
08:55:11 <LordAro> hang on, that rings a bell...
08:55:20 <ahyangyi> I want to pretend I don't understand, but I run Gentoo Linux so I have to.
08:56:29 <fairyflossy> So I'll use myself as an example here. I'm a programmer, I studied how to write programs before AI became the thing it is now. If I wanted to make a program that did a thing, I had to either write it myself, or look up code from other people to then work off of as a base. In either option, I got more involved in the code myself, it was up to me to take these pieces and fit them together, tweak
08:56:29 <fairyflossy> them here and there, constantly going back and forth to make modifications. Even if I didn't fully understand what was going on at all times, I at least had some familiarity with the code, and got an idea for how it all interacted.
08:56:29 <fairyflossy> Now, if I go to say, ChatGPT and tell it "Hey, I want a program that does this", it'll spit out a load of code for me. Sure, in theory I could go read through it and note for myself "okay, this does X, this does Y", but in practice I don't think people would do that. So now, I have a chunk of code that does what I want it to do, and if it works... Well, then that's that. I have a program. How does
08:56:29 <fairyflossy> this program work? I dunno. Did I figure out any tricks or useful functions for future projects? No. I didn't learn anything from it.
08:57:40 <dh1> yeah, that's all fine, but actually 90% of the time i want to automate something i don't need to learn anything, i just need my files renamed
08:58:18 <fairyflossy> And I think that's fine, but the issue comes from when people want to do more with it than just "Simple thing"
08:58:20 <dh1> (which obv this is just general rambling and not necessarily relevant to openttd policy stuff)
08:59:05 <dh1> yeah, and so i think probably the world is gonna shamble around a bit and eventually settle on some sort of llm equilibrium
08:59:30 <fairyflossy> There's a difference between "I use ChatGPT to do a thing to speed up something" versus "I'm going to integrate ChatGPT into my career"
08:59:43 <dh1> but in the meantime we are gonna have people being insufferable on linkedin
09:00:02 <fairyflossy> But is that anything new?
09:00:18 <dh1> they can be insufferable quicker
09:01:30 <dh1> As an Insufferable LinkedIn User I want to...
09:02:48 <dh1> all of this channels logs are public and presumably have been ingested into multiple llms too
09:03:18 <andythenorth> we can't incorporate unmodified LLM generated code from the major LLM vendors
09:03:46 <dh1> i think it is going to be interesting in 20 years (if we still have society) to see how language has become more esoteric and weird and inconsistent as people try to distinguish themselves from chatbots
09:03:49 <andythenorth> the author can't assert copyright, or a GPL compatible alternative license, so it doesn't comply with GPL
09:04:11 <andythenorth> the debate about the rest of it is all very interesting
09:04:29 <andythenorth> but we have a contribution framework based on copyright
09:04:30 <LordAro> IMO copyright was already dead, LLMs just set fire to the corpse
09:05:01 <andythenorth> it's a whole other debate about whether copyright is actually an unequivocal moral good (as it's assumed to be)
09:05:35 <andythenorth> or whether that's just the result of a lot of public lobbying by MPAA, CPS, PRS etc
09:05:35 <ahyangyi> Anyways, I would not worry about StackOverflow-sized code snippets.
09:05:39 <dh1> we did a library of babel thought experiment at law school where we autogenerated a stupid number of things and had to argue copyright interest in the outputs
09:05:58 <dh1> fuck knows where i landed on that at the time because i too was an insufferable linkedin user and law student
09:06:01 <andythenorth> did the answer vary by jurisdiction and established case law?
09:06:02 <ahyangyi> "Hey, you violated GPTv2 because you copied that one line from a StackOverflow answer you visited"
09:06:16 <dh1> what jurisdiction is openttd in
09:06:25 <andythenorth> unproven, the case has not come to court yet
09:07:49 <LordAro> UK registered company
09:08:16 <LordAro> Whether or not that covers the source code rather than all the certificates/websites/etc though...
09:10:20 <andythenorth> possibly the output of a foundation model trained only on GPL-compatible sources might be valid
09:10:40 <dh1> i'm not sure copyright law looks behind the veil of the medium
09:11:00 <andythenorth> armchair lawyering tends to be unwise, and engineers and programmers tend to think legal cases are about clear fixed rules, not argumentation
09:11:04 <dh1> this is one of the "problems" (for want of a better word) with using law to create fake things
09:11:18 <dh1> you create an abstraction that is incredibly hard to reconcile with the real world
09:11:33 <andythenorth> because there's no artefact chain?
09:11:54 <ahyangyi> Anyways, there are unsettled questions and it's wise to err on the side of caution
09:11:54 <dh1> in nz law at least the statutory position is that if a work is computer-generated it's author is the person who arranged for it to be generated
09:12:14 <andythenorth> similar in UK, there's case law
09:12:19 <ahyangyi> Even if I personally don't believe LLM outputs are encumbered with any kind of license issues.
09:12:26 <andythenorth> there are specifics about 'caused to be generated'
09:12:30 <dh1> yeah i assume our copyright act is closely related to uk
09:12:46 <andythenorth> a database retrieval isn't a generation
09:13:04 <andythenorth> but a world created in a sandbox computer game is
09:13:35 <ahyangyi> And one can (partially) reconstruct the training material from the LLM weight, so one might argue that it is actually a database retrieval in disguise.
09:13:45 <ahyangyi> Not an argument that I would make though.
09:14:10 <andythenorth> typicallt the random seed makes it more stochastic
09:15:15 <dh1> on the wording of the nz act at least i reckon it would hang on the meaning of computer-generated, because a plain english reading seems to suggest that the copyright in the original source material is irrelevant
09:15:18 *** Flygon has quit IRC (Read error: Connection reset by peer)
09:16:12 <andythenorth> until a plaintiff or classs of plaintiffs win a case establishing alternative legal facts
09:16:22 <andythenorth> like the Anthropic case...
09:17:08 <dh1> it's usually pretty hard for a uk or nz judge to override the plain meaning of a statute
09:17:25 <andythenorth> yes but punitive tariffs
09:17:47 <andythenorth> dunno, this all escalates ridiculously 😛
09:18:22 <dh1> sure, but that's a political consideration
09:21:39 <dh1> anyway i need to finish this feature request for microtransactions
09:25:04 *** dh1 has quit IRC (Quit: My Mac has gone to sleep. ZZZzzz…)
09:40:09 *** Smedles has joined #openttd
11:20:01 *** ufo-piloot has quit IRC (Ping timeout: 480 seconds)
11:56:52 *** ufo-piloot has joined #openttd
13:09:23 *** Smedles has joined #openttd
14:02:28 *** lobstarooo has quit IRC (Read error: Connection reset by peer)
14:06:57 *** lobster has joined #openttd
14:32:43 <peter1138> LordAro, I have C# dependency issues, sad :(
14:33:37 <LordAro> peter1138: we have some java dependency issues
14:34:03 <LordAro> seems maven doesn't send auth if it's asking for package metadata (because of a dependency range)
14:34:19 <peter1138> This is just design issues... B depends on A, now I need something from B in A
14:37:06 *** Wormnest has joined #openttd
14:44:21 <peter1138> Oh ffs, and then there's the models that need to be split out.
14:44:32 <peter1138> Shove it all in one library, did you say...
16:43:09 *** gelignite has joined #openttd
17:06:36 <peter1138> Still not fixed it.
17:32:55 <yiffgirl> ahyangyi: the last three were just in general, sorry for giving the impression it was targeting you specifically. but moving your [multi] cursor is something that's been around since, what, vi?
17:32:55 <yiffgirl> ...unless the llm does that without any user input whatsoever. which sounds like hell to me, but would be something classical tools can't do
17:33:57 <ahyangyi> Moving your cursor according to rules, yes
17:34:07 <ahyangyi> Moving your cursor according to statistics, no.
17:36:16 <peter1138> Moving my cursor based on what it thinks I might want to edit sounds dire.
17:36:55 <LordAro> i have previously wanted a "focus follows intent" for tiling window managers
17:37:06 <LordAro> but usually that's actually "focus follows eyes"
17:37:16 <ahyangyi> I'm still a good old vim coder by the way
17:38:16 <yiffgirl> wish i could get into modal editing more. the only thing i really know is :wq
17:39:19 <peter1138> I used to do all my OpenTTD development with vim, back before there was much in the way of helpers.
17:40:13 <peter1138> So I never figured out how to configure it for a more advanced development environment with ctags and language servers and what not.
17:40:25 <yiffgirl> kate is my editor of choice these days
17:41:09 <yiffgirl> now given, it doesn't play nice with the openttd codebase for whatever reason, but it works well enough for me
17:41:10 <peter1138> (Although I still use VS Code for $work related .Net stuff as parts of that are not available in Codium.)
17:41:20 <LordAro> Still vim for me. Never seen the need for neovim
17:41:40 <yiffgirl> LordAro: have you tried helix ? =p
17:42:19 <peter1138> Thinks like ctrl-clicking on a function call to take you directly there, in another file... I have no idea what the vim incantation would be for that.
17:42:53 <ahyangyi> ctrl+], but some prior set up is needed
17:42:55 <LordAro> (but mostly because that would require trying a new editor, not because of the AI stuff)
17:43:43 <LordAro> git grep is perfectly sufficient :p
17:44:35 <yiffgirl> ahyangyi: a modal terminal editor in rust
17:44:35 <yiffgirl> has different bindings from vim though
17:44:46 <yiffgirl> oh oops wrong reply sorry
17:46:04 <_jgr_> I haven't had great experiences with fancy IDEs, it seems like I spend more time debugging the editor and project files than actually using them
17:46:47 <_jgr_> I use geany which is pretty dumb but mostly frustration free
17:47:38 <yiffgirl> oh, i used that for a long while before switching to kate
17:49:13 <peter1138> Oh, I see, more OS-specific functionality gated behind #define.
18:00:24 <andythenorth> did I miss OpenTTD 15 while I was on holiday? 👀
18:22:06 <andythenorth> before I get even more confused....
18:22:14 <andythenorth> the nml railtype table "fallbacks"
18:22:21 <andythenorth> they're nothing to do with the grf spec?
18:22:41 <andythenorth> they're a mistake that has been cargo-culted around multiple grfs?
18:24:17 <andythenorth> RT_SHINY_RAIL: [SHNY, RAIL]
18:24:19 <andythenorth> I can't see anything reference to the fallback list in the actual grf spec for railtypetable
18:59:02 <_glx_> It's kinda magic using act7/9 IIRC
19:12:09 *** brickblock19280 has joined #openttd
19:12:09 <brickblock19280> Yeah some action D too iirc
19:13:06 *** Wormnest has quit IRC (Ping timeout: 480 seconds)
19:14:29 <andythenorth> I suspect it was added to ease compatibility, but it was unfortunate
19:15:01 <andythenorth> it has been widely cargo culted, and breaks proper domain for railtypes
19:15:31 <andythenorth> afaict, railtype compatibility was deliberately designed to be additive
19:16:17 <andythenorth> this fallback system makes it exclusionary, and makes it trivial for one grf to 'break' another grf, but only when a third grf is added
19:17:29 <andythenorth> I guess it is just what it is 🙂
19:20:16 *** WormnestAndroid has quit IRC (Ping timeout: 480 seconds)
19:20:24 *** WormnestAndroid has joined #openttd
19:21:06 <_glx_> basically for `RT_SHINY_RAIL: [SHNY, RAIL]` the code does if SHNY is defined set a temp param to SHNY (param default to RAIL), then the param value is used in act6 to modify the actual railtype act0
19:24:02 <_glx_> and it does that starting from the end of the list, so the param will contain the first defined railtype from begin of the list
19:27:01 <andythenorth> as vehicle authors seem to widely believe the fallback table is the method to specify compatibility
19:27:10 <_glx_> it moves track compatibility to vehicles
19:27:22 <andythenorth> however the legacy fleet of existing grfs will be unremediatable
19:28:09 <_glx_> `railtypetable` is just a list of railtype expected by your grf
19:31:54 <_glx_> with a mechanism to dynamically fill it when loading the grf based on other grfs
19:41:14 *** zsombi55 has quit IRC (Quit: User went offline on Discord a while ago)
19:53:12 <peter1138> I'm gonna need a Windows PC at some point :(
19:58:15 <_glx_> anything I can help with?
20:09:08 <peter1138> > 21 files changed, 2 insertions(+), 627 deletions(-)
20:35:28 *** toktik has quit IRC (Remote host closed the connection)
21:26:54 *** dh1 has quit IRC (Quit: Textual IRC Client: www.textualapp.com)
21:40:46 *** Wolf01 has quit IRC (Quit: Once again the world is quick to bury me.)
21:54:35 *** keikoz has quit IRC (Ping timeout: 480 seconds)
22:08:06 *** WormnestAndroid has quit IRC (Read error: Connection reset by peer)
22:08:35 *** WormnestAndroid has joined #openttd
22:24:40 *** Wormnest has joined #openttd
22:53:41 <jfkuayue> My home printer can only be used on windows...
continue to next day ⏵
