00:12 <+bridge> glad you can solve that issue, hope you can also tackle blocking prs by not answering, gameplay feature stuff, and trying to revert everything mentality 00:17 <+bridge> i have one question. i have own maps they're not shows me a flags how i can enable it? 00:18 <+bridge> i with server testing 00:30 <+bridge> hey dev? 00:33 <+bridge> Lots of dreams 00:34 <+bridge> Lots of dramas 00:34 <+bridge> how it doing? 00:35 <+bridge> omg join girl in cam discord.gg/xQHpr8aq 00:36 <+bridge> What 00:36 <+bridge> I don't know what's your issue 00:40 <+bridge> ayyy we got the other guy here 00:41 <+bridge> https://cdn.discordapp.com/attachments/1362129113950851104/1364891564312100924/C91E7AB1C07FB881B26E237B9211D038.gif 00:41 <+bridge> gadze :3 00:42 <+bridge> was my test server not shows a flag with my map 00:42 <+bridge> i have a it paste 00:43 <+bridge> What do you mean flag 00:43 <+bridge> like flag ingame sec showing 00:44 <+bridge> here a big problem or just bug 00:44 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1495193235507642539/2085CD1.JPG?ex=69e55ab6&is=69e40936&hm=7e0cfbade83f99a1e129ffca9e8ee41dde93f5832bcc583e243c78791922599d& 00:44 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1495193235939791039/2041181.JPG?ex=69e55ab6&is=69e40936&hm=e399834ea51ec52a49ec41dbdd1f34d23fe2c33fb21802801622083b6dbf5ab7& 00:44 <+bridge> dont shows me a flag 0-0 00:44 <+bridge> who can help me how this it enable? 00:45 <+bridge> @pioooooo answer me? 00:45 <+bridge> you need a mod for that 00:45 <+bridge> just ddnet server for me. i soon own will 00:46 <+bridge> flag does not work in normal ddnet server 00:46 <+bridge> you need something like F-DDrace 00:46 <+bridge> 0-0 i soon will own doing. mabye doing for ddnet server allow only flag for test server? 00:47 <+bridge> no, ddnet does not have flag 00:48 <+bridge> there is not flag code in ddnet server 00:48 <+bridge> thats why you need a mod 00:48 <+bridge> i need code this a trun on? 00:48 <+bridge> like i mentioned 00:48 <+bridge> you need F-DDrace 00:48 <+bridge> i dont need f-ddrace i will doing own code 00:48 <+bridge> or DDNet++ 00:49 <+bridge> sure 00:49 <+bridge> Then you have to code it yourself 00:49 <+bridge> https://tenor.com/view/shake-head-gif-24919122 04:15 <+bridge> I need help solving a problem with parsing logs, because this looks like a limitation: 04:15 <+bridge> 04:15 <+bridge> There are 2 players, `player 1` and `player 1: not`. 04:15 <+bridge> The logs look like: 04:15 <+bridge> ``` 04:15 <+bridge> 2026-04-19 03:06:31 I chat/all: player 1: hi 04:15 <+bridge> 2026-04-19 03:06:35 I chat/all: player 1: not: hi 04:15 <+bridge> ``` 04:15 <+bridge> We know the first log is `player 1` saying `hi`. 04:15 <+bridge> How do we disambiguate the second log, between: 04:15 <+bridge> - `player 1` saying `not: hi`; 04:15 <+bridge> - `player 1: not` saying `hi`? 04:16 <+bridge> looks non-distinguishable 04:17 <+bridge> but that's also the client. you can modify it instead of parsing the logs 04:17 <+bridge> <0xdeen> Why do you want to parse the logs in the first place? The format is not great for that as you just found out 04:18 <+bridge> My current tools are using `cl_input_fifo` and `logfile` so that I can have custom logic without the need of having a custom client. 04:19 <+bridge> I read what's happening using `logfile` and decide what the client does with `cl_input_fifo`. 04:20 <+bridge> non-distinguishable, you could theoretically maintain a list of players via join/leave events and split at those as ground truth 04:20 <+bridge> I thought of that, but then the player 1 can't say `not: 1` because it will look like the 2nd case. 04:20 <+bridge> I thought of that, but then the player 1 can't say `not: hi` because it will look like the 2nd case. 04:20 <+bridge> unlikely that this happens, you could mark the line as ambiguous if you need data correctness 04:22 <+bridge> I'm more worried about players trying to make other players say something they didn't. I could do that and keep a list of players, but then it means I can't confidently parse past logs. 04:24 <+bridge> tbh, if you want to go so far i'd just patch the client ATP, you're pretty much at the ceiling of what the structured output can give ya xd 04:24 <+bridge> <0xdeen> Create a json output format instead, where each line is an event 04:25 <+bridge> or sniff the network and get the chat messages from there 04:25 <+bridge> All these problems happen because we allow `"`, `'` and `:` characters to be used in nicknames without escaping them in all logs. The same thing happened with [#11879](https://github.com/ddnet/ddnet/issues/11879). 04:25 <+bridge> https://github.com/ddnet/ddnet/issues/11879 04:25 <+bridge> -# *There was no need for that* 04:26 <+bridge> I would like that, actually. That's already what I'm making my tool do. 04:28 <+bridge> The problem is that the source itself is what's making it ambiguous when making these logs. 04:30 <+bridge> tbf the logs are more made to be read, rather than parsed 04:30 <+bridge> (which is probably true of most programs) 04:33 <+bridge> Yeah, I get that, but I wonder what this means for when an admin or moderator needs to check the logs when proving someone guilty. Someone could easily impersonate them with this. 04:33 <+bridge> <0xdeen> yeah, I meant the json mode should be optional, for people who want a tool to parse it 04:36 <+bridge> Luckly server-side logs, like the rcon command `dump_log`, don't suffer from this because the name is logged separate from the message, but it means they can't trust the client-side logs they have when checking what was said while they were connected. 04:39 <+bridge> You could add the id of the player at the start of the message, so it would look like: 04:39 <+bridge> ``` 04:39 <+bridge> 2026-04-19 03:06:31 I chat/all: 03 - player 1: hi 04:39 <+bridge> 2026-04-19 03:06:35 I chat/all: 24 - player 1: not: hi 04:39 <+bridge> ``` 04:39 <+bridge> It wouldn't solve the problem without having the player list at hand, but it would be enough to at least see that something is wrong. 04:39 <+bridge> _ _ 04:39 <+bridge> You could add the id of the player at the start of the message, so it would look like: 04:40 <+bridge> ``` 04:40 <+bridge> 2026-04-19 03:06:31 I chat/all: 03 - player 1: hi 04:40 <+bridge> 2026-04-19 03:06:35 I chat/all: 24 - player 1: not: hi 04:40 <+bridge> ``` 04:40 <+bridge> It wouldn't solve the problem without having the player list at hand, but it would be enough to at least see that something is wrong. 12:15 <+bridge> 19.8.1 seems to be blocked easily by smart app control on winshit 12:15 <+bridge> can't run ddnet anymore tho 12:17 <+bridge> Add an exception, nothing we can do about it 12:19 <+bridge> smart app control seems to be shit (even tho I'll probably not disable it), you can't even put an exception 12:20 <+bridge> and apparently to fix it you need to perform code signing 12:20 <+bridge> it's basically "we haven't seen the executable before" 12:20 <+bridge> "need to pay up" 12:20 <+bridge> ye just checked it's AI based :KEKSad: 12:20 <+bridge> https://klipy.com/gifs/programming-code 12:21 <+bridge> imagine using windows 12:21 <+bridge> you can disable it to send a signal to microsoft that not having exceptions isn't cool 12:21 <+bridge> "our numbers show that users don't disable it" 12:21 <+bridge> damn I really hope these idiots stop thinking about windows only and makes their anticheat Linux compatible. I'll beat the WR of uninstalling windows after that 12:21 <+bridge> Doubt they'll listen 12:21 <+bridge> > There is currently no way to bypass Smart App Control protection for individual apps. You can turn Smart App Control off, or (better yet), contact the developer of the app and encourage them to sign their app with a valid signature. 12:22 <+bridge> damn I really hope these idiots stop thinking about windows and make their anticheat Linux compatible. I'll beat the WR of uninstalling windows after that 12:23 <+bridge> If it sees poor adoption then it wouldn't be economical to keep maintaining it 12:23 <+bridge> True that 12:23 <+bridge> even more if it's AI based, but 90% of users doesn't understand anything and will just keep it enabled as it is on by default 12:23 <+bridge> even more if it's AI based, but 90% of users don't understand anything and will just keep it enabled as it is on by default 12:24 <+bridge> alright I'll disable it 12:24 <+bridge> Btw Rei why did you block me after running to multiple other people to blackmail me, after you talked down on me personally and my mod? 12:24 <+bridge> take that to DMs 12:24 <+bridge> cant 12:24 <+bridge> I wanted to earlier 12:24 <+bridge> don't do it here 12:24 <+bridge> Ok 13:00 <+bridge> is that the first indication that the rust rewrite is not more robust? 13:00 <+bridge> As I remember after every release, we probably should start signing these 13:01 <+bridge> I wanted to review the PR, then I saw it is Rust :( 13:02 <+bridge> I'd say it's more the 0.7 bridge part that is not exactly perfectly sane 13:02 <+bridge> k, hence the question. 13:03 <+bridge> who's able to review the rust prs? 13:04 <+bridge> heinrich, robyt3, learath 13:04 <+bridge> and some others maybe 13:04 <+bridge> ok. at least some people. 13:05 <+bridge> I did not keep up when they started learning rust in here a few years ago. I knew the gap would be too great too soon 13:05 <+bridge> Full disclosure, I've never been a huge fan of this sort of mixing of Rust and C++ 13:06 <+bridge> i'm just wondering long term 13:06 <+bridge> Mixing languages except for adding C seems to always be far more complicated than desired 13:06 <+bridge> heinrich, god forbid, getting hit by a bus and then nobody being able to understand any of that code at all. 13:07 <+bridge> Look at the amount of code cxxbridge generates? Technically all of that is attack surface, attack surface that is almost impossible to audit that we are just hoping dtolnay got right 13:07 <+bridge> Look at the amount of code cxxbridge generates. Technically all of that is attack surface, attack surface that is almost impossible to audit that we are just hoping dtolnay got right 13:08 <+bridge> Whenever mixing languages I highly encourage exporting a C API. Does it look ugly? Yes, but it's also is very easy to reason about and audit 13:09 <+bridge> This is very true and I think everyone has an example of this in their company if it's not a startup 13:10 <+bridge> What do you guys think should be done about the old releases on the downloads page? They are technically now all vulnerable 13:11 <+bridge> archive + warning 13:11 <+bridge> what's the issue? 13:13 <+bridge> i'm giving an answer without having read the problem D: better ask late than never 👀🔥. 13:13 <+bridge> If we still want to make the old versions available at all, replace the download links with a wrapper page that first asks for confirmation, I guess. 13:14 <+bridge> Does the master server know the client version? Can't we make the list filter all non-community servers, possibly all servers except DDNet? 13:14 <+bridge> RCE likely possible if you connect to a malicious server 13:14 <+bridge> holy 13:14 <+bridge> Not easy, but possible-ish 13:15 <+bridge> "likely"? that sounds way more possible than what you responded yesterday :o 13:15 <+bridge> > Yeah, in this case we evaluated the "potentially" to be very hard in practice, so it would not have been the end of the world 13:16 <+bridge> what is the core cause of the issue, out of interest? 13:16 <+bridge> OOB 13:16 <+bridge> i think 13:17 <+bridge> out of bounds 👀 hmmm. 13:18 <+bridge> is backporting a fix feasible? 13:19 <+bridge> That would probably require a simpler c++ fix 13:19 <+bridge> Not all versions support Rustt 13:19 <+bridge> Not all versions support Rust 13:19 <+bridge> i have not seen the fix, just wondering if the fix is easy enough to be backported. 13:20 <+bridge> I mean backporting to a few major versions, not all versions since dawn. 13:22 <+bridge> i wonder if it's possible to analyze, what versions the majority of players use. 13:23 <+bridge> This is likely what we'll be shipping for a 19.8.2 since the Rust one seems to have more edge issues than I anticipated. I want it to at the very least cook a little longer in nightly or perhaps be replaced by a larger C++ rework of the snap code 13:23 <+bridge> btw dtolnay works at palantir 13:24 <+bridge> There was something that I'd forgotten about it that makes it easier than I remember. Still would take a very crafty attacker, but there is one subtle detail that makes it easier than I remembered last night talking to you 13:25 <+bridge> that's a good one. 13:25 <+bridge> That guy is definitely not real, just a wizard 13:25 <+bridge> One of the craftiest programmers I have ever seen 13:26 <+bridge> defense probably pays well 13:26 <+bridge> I did consider this for a bit, but we don't really have a mechanism to push an update like that to people, or warn people on older versions reliably 13:27 <+bridge> It would make the stuff on the downloads page safe again though 13:27 <+bridge> yeah he is a magician 13:27 <+bridge> btw he made a pr to a rust crate i maintain 13:27 <+bridge> im honored 13:27 <+bridge> at least that part, where new downloads are not exploitable 13:27 <+bridge> Which one? 13:27 <+bridge> https://github.com/mlir-rs/tblgen-rs 13:28 <+bridge> https://github.com/mlir-rs/tblgen-rs/pull/54 13:28 <+bridge> Ah, LLVM stuff 13:30 <+bridge> We have ingame news to inform them. Based on checking the version in the request header, the news for old client versions could always show a security alert. We could also consider adding a feature to new versions to show an mandatory alert when launching the client based on a message provided in the ddnet-info to handle this in the future. 13:54 <+bridge> XD 13:54 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1495392049774460989/image.png?ex=69e613df&is=69e4c25f&hm=d197bd29e1c9317175389508335c8d62faf4c9fd1d605cfe92e8fccbb78a337f& 13:54 <+bridge> good review 👍 13:57 <+bridge> Thanks 13:58 <+bridge> yeah its more of a quick bandaid rather than a fix, but if it works, it works :Shruggo: 14:15 <+bridge> > but if it works, it works 14:15 <+bridge> That's not true for many other unmerged/unreviewed PRs tho 14:16 <+bridge> that is true 14:21 <+bridge> There is a difference between "fixes a bug that instantly crashes the client" vs other things, them getting treated differently shouldn't be too much of a shocker 14:24 <+bridge> helo fokko 14:24 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1495399770116722858/image.png?ex=69e61b10&is=69e4c990&hm=e7edfb5f3742fba69b53ce79c23a0aab531ad4b4e9f0fee4f8607107536e69fa& 14:25 <+bridge> my dummi 14:25 <+bridge> foo bar baz 14:25 <+bridge> Sec 14:26 <+bridge> Which client are you using? 14:26 <+bridge> chillerbot-ux 14:27 <+bridge> I'll release you, but I'm not sure this is fixable in a good way 14:27 <+bridge> oki thanks! 14:27 <+bridge> Try please, and let me know if issues arise 14:30 <+bridge> with rust change any malicious server can just crash anyones game and theres no crash log or anything 14:30 <+bridge> i guess ur rust isnt miracle that fixes everything if u dont know how to code 14:36 <+bridge> https://tenor.com/view/supa-hot-fire-black-guys-roasted-take-that-black-gif-4954520 14:38 <+bridge> Funny how everytime you guys do this you get confused as to why you get timeouts and stuff. Surely you see how this is just inflammatory and personal 14:38 <+bridge> I did not check the PR further, but I can assume crashing the client without a log is actually bad 14:39 <+bridge> no problem, i wont disclose the bug and keep it secret 14:39 <+bridge> hope you can figure out a fix in secret from community too 14:39 <+bridge> :p 14:40 <+bridge> 1sec let me record vid so u know is real : 14:40 <+bridge> 1sec let me record vid so u know is real 14:40 <+bridge> There are dozens of server -> client crash bugs that are non-exploitable. You can ask your AI 14:41 <+bridge> In this specific case I do not feel mentioned 14:41 <+bridge> and rust introduced dozen more 14:46 <+bridge> 😭 14:46 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1495405200070934608/2026-04-19_14-46-38.mp4?ex=69e6201e&is=69e4ce9e&hm=779efa9cb62fa393c908bb19fdbd4d3fae593d57793fe58ebcbce24d7b180b6a& 14:49 <+bridge> yay 14:52 <+bridge> God that CI pipeline has grown extremely slow 14:55 <+bridge> Whole DDNet building is slow 14:55 <+bridge> Try and compile F-DDrace, it's super fast (server only ofc= 14:55 <+bridge> Try and compile F-DDrace, it's super fast (server only ofc) 14:56 <+bridge> Is it the rust dependencies? 14:56 <+bridge> Those are cached iirc 14:57 <+bridge> Maybe CodeQL? Idk I'll take a look 14:57 <+bridge> What if you git blame? 14:57 <+bridge> I remember git blame is horrible on ddnet building times 14:57 <+bridge> do you run clang-tidy? 14:59 <+bridge> It seems clang-tidy, ubuntu-latest-fancy and windows-latest-mingw take the longest 15:00 <+bridge> Ah I guess it's code coverage that takes a long time 15:02 <+bridge> Still doesn't add up to the time it took, maybe we don't get nodes quick enough because we are on the free tier? 15:04 <+bridge> Either way, it seems unlikely we can get it below 30 minutes no matter what given that's how long clang-tidy takes 15:06 <+bridge> 30 minutes 😭 15:06 <+bridge> my ci takes 40-60 minutes depending on how the stars align 15:08 <+bridge> No 15:37 <+bridge> I also get really annoyed by this tbh, maybe I take a time off of discord 15:40 <+bridge> rly why 15:40 <+bridge> i just searche this guy up 15:41 <+bridge> surely u can get a job at a less controversial company with all this clout 15:53 <+bridge> <01000111g> Fo i have to assign labels to my issue myself, or will the maintainers do it? I did not see the option to 15:53 <+bridge> <01000111g> Do i have to assign labels to my issue myself, or will the maintainers do it? I did not see the option to 15:55 <+bridge> only maintainer 15:56 <+bridge> i doubt anywhere else pays like that 16:03 <+bridge> damn I wonder how they are doing this 16:03 <+bridge> I can't start DDNet while Battlefield 6 is running 16:03 <+bridge> I can't start DDNet while Battlefield 6 is running, it instantly kills it 16:06 <+bridge> Which kills which? 16:12 <+bridge> either we call a custom panic hook (i.e windows::Win32::System::Diagnostics::Debug::RaiseException), or we just abort(); which should generate an unhandled exception that **can** be catched, i think. 16:12 <+bridge> either we call a custom panic hook (i.e `windows::Win32::System::Diagnostics::Debug::RaiseException`), or we just abort(); which should generate an unhandled exception that **can** be catched, i think. 16:14 <+bridge> BF6 kills DDNet 16:14 <+bridge> Huh? 16:14 <+bridge> interesting. xd 16:14 <+bridge> I have thought it's BF6 anti-cheat. 16:15 <+bridge> Can you run DDNet with gdb, or check which exitcode it returns? 16:19 <+bridge> yup give me some time but I'm pretty sure it's not intended 16:19 <+bridge> nevermind it seems intended 16:19 <+bridge> I assume BF just sends a quit window message to DDNet though 16:19 <+bridge> I don't even get an exit code on windows 16:20 <+bridge> You can get the exit code by running the client in `cmd`, then execute `echo %errorlevel%` as the next command 16:22 <+bridge> echo %errorlevel% 16:22 <+bridge> 9009 16:25 <+bridge> Weird. If you can, try to get a backtrace with gdb. 16:25 <+bridge> I don't have mingw setup on this machine but msvc 16:26 <+bridge> ``` 16:26 <+bridge> Exception non gérée à 0x00007FFFB6120D39 (nvoglv64.dll) dans DDNet.exe : 0xC00000FD: Stack overflow (paramètres : 0x0000000000000001, 0x0000007DD4B03FF8). 16:26 <+bridge> ``` 16:26 <+bridge> can try to set it up 16:26 <+bridge> Seems like it crashes the NVIDIA driver 16:27 <+bridge> fun fact if you run DDNet before BF6 it doesn't kill ddnet 16:27 <+bridge> it's only when you start DDNet after BF6 16:27 <+bridge> Did you try if Vulkan/OpenGL are different? 16:28 <+bridge> yeah it was my guess at https://discord.com/channels/252358080522747904/293493549758939136/1495428516576104810 16:28 <+bridge> but didn't work 16:29 <+bridge> Maybe BF6 just allocates all your VRAM ahead of time. Or it's an NVIDIA driver bug. 16:29 <+bridge> feature is accepted btw, this means we'd allow a PR for it. It's also marked as "good for beginners" which are issues for new people who want to work themself into the project, so they get like an onboarding. It just wasn't picked up for half a year now 16:29 <+bridge> ah probably yes 16:35 <+bridge> good morning fellow ddevs 17:59 <+bridge> Holy llm 18:00 <+bridge> HAHAHAH 18:02 <+bridge> doesnt look like it wdym 18:02 <+bridge> i mean there is the — character 18:02 <+bridge> The pr description 18:03 <+bridge> > Not fatal on modern hardware, but a 6× regression on a hot path is worth investigating before it compounds with future changes. 18:03 <+bridge> 18:03 <+bridge> bro typed 6× instead of 6x 18:03 <+bridge> is that normal 18:03 <+bridge> true, doesnt mean the benchmark isn't correct 18:03 <+bridge> yes. 6x kinda crazy 18:04 <+bridge> yea even if its llm written the concerns are still valid 18:04 <+bridge> However idk why ppl just copy paste whatever the llm outputs. Just summarize it in 5 sentences, then if you want all the detail put it somewhere else so ppl dont have to waste time reading unnecessary info 18:05 <+bridge> too much effort 18:05 <+bridge> check him pc this is not normally 18:05 <+bridge> People throw it onto everything they dislike nowadays 18:06 <+bridge> Lol 18:06 <+bridge> what 18:07 <+bridge> i mean there are signs of llm usage 18:07 <+bridge> it's a meme reference, no timeout pls 18:07 <+bridge> <_laveer> description is clearly ai generated, or bloated, but afaik that guy is 0xf dev, so I assume he's one of the guys who is should be concerned by performance issues 18:07 <+bridge> Did the LLM do the benchmark? Did the LLM come up with made up values? 18:07 <+bridge> <_laveer> description is clearly ai generated, or bloated, but afaik that guy is 0xf dev, so I assume he's one of the guys who should be concerned by performance issues 18:08 <+bridge> well i dont know that and i do agree even if its llm written the concerns are still valid 18:09 <+bridge> the bechnmark file is included so I guess someone could check it out 18:09 <+bridge> but im also worried from what ive seen before that heinrich wont interact if they are using a LLM to communicate 18:10 <+bridge> like on 11903 18:10 <+bridge> #11903 18:10 <+bridge> https://github.com/ddnet/ddnet/issues/11903 18:11 <+bridge> Theres no provlem with using an Llm to do the benchmark. But I dont think ppl should copypaste llm output straight into the pr description. 18:12 <+bridge> it's something different here. nobody is presenting their cool code or whatever. this is a real and valid concern, possibly processed with the help of ai as a tool. heinrich cant label this as bad or mean towards the reader. he pushed thousands of lines of rust cod NO ONE can read. 18:12 <+bridge> it's something different here. nobody is presenting their cool code or whatever. this is a real and valid concern, possibly processed with the help of ai as a tool. heinrich cant label this as bad or mean towards the reader. he pushed thousands of lines of rust code NO ONE can read. 18:13 <+bridge> Using LLM is a valid choice here, i would've done the same ( for the benchmarking, i mean - cba to write it myself ) - but i think its valid to at least write it in your own words and translate it using https://deepl.com or similar tools - communicating with a machine is weird IMO 18:13 <+bridge> Using LLM is a valid choice here, i would've done the same ( for the benchmarking, i mean - cba to write it myself ) - but i think its valid to request to at least write it in your own words and translate it using https://deepl.com or similar tools - communicating with a machine is weird IMO 18:14 <+bridge> DeepL AI Platform 18:14 <+bridge> even google uses AI/LLM's to translate :p 18:15 <+bridge> so what is the difference between using a translator and a llm? 18:15 <+bridge> <0xfaulty> I think the LLM actually did a pretty good job of summarizing the benchmark results and splitting them into sections - better than my own first pass, to be honest. Anyway, as already mentioned above, anyone can verify this on their own using the test file, or point out if I got something wrong. There is no need for any major changes to check it, just add the test file and hook it up in CMake. And if you have concerns about the benchmark code, or s 18:16 <+bridge> one talks for you, the other recites what you wanted to say in a different language, i dont see how you can treat them as equal 18:18 <+bridge> true but you dont know how he might of used it he might of written in his most fluent language then asked the llm to translate it which may be more accurate than using a translator 18:20 <+bridge> I don't see why translating using LLM is considered as bad. As long as you understand the generated sentence and agree it has nothing more than what you wanted to say, it's fine 18:21 <+bridge> that would just make it worse 18:21 <+bridge> also this atleast gives quite lot of detail 18:21 <+bridge> even if its llm its better than ur average report 18:22 <+bridge> this just shows this was rushed in a bad way 18:22 <+bridge> by ddnet side i mean 18:22 <+bridge> I prefer a LLM that will also fix my mistakes in my sentence than a translator that has higher chance to generate bad sentences behind 18:23 <+bridge> i'm not disagreeing with the choice of LLM's at all, i'm simply stating what heinrich requested in a few of those PR's 18:23 <+bridge> I prefer a LLM that will also fix my mistakes in my sentence than a translator that has higher chance to generate bad sentences behind 18:23 <+bridge> (It helps to learn aswell) 18:23 <+bridge> i.e "i prefer human-written words" 18:23 <+bridge> thats hein preference 18:23 <+bridge> does he represent the entirety of ddnet 18:23 <+bridge> ? 18:23 <+bridge> i prefer this detailed issue 18:23 <+bridge> do i? ._: 18:23 <+bridge> i dont like reading llm output either 18:23 <+bridge> ur justb ikeshedding on real issues 18:24 <+bridge> I made a one-off comment and now ppl are making it bigger than what it needs to be 18:25 <+bridge> Yeah for once I kinda like the LLM output 18:27 <+bridge> I dont mind him using AI, which i also said. :/ 18:28 <+bridge> shocker rust doesnt miracously make the code faster either XD 18:28 <+bridge> XD 18:28 <+bridge> 🤦 18:29 <+bridge> Funny 18:31 <+bridge> put in another 10k unreviewd loc 18:31 <+bridge> that'll fix it 18:31 <+bridge> So true 18:31 <+bridge> im shocked its 6x 18:31 <+bridge> xD 18:31 <+bridge> Rust is too safe 18:32 <+bridge> learath can u give me a fact about crabs 18:33 <+bridge> <_laveer> rewrite that in zig now, should work 18:33 <+bridge> No facts in #developer 18:33 <+bridge> zig would actually be a valid choice 18:33 <+bridge> zig would segfault 18:33 <+bridge> since it can natively interpret C 18:34 <+bridge> is it also so safe so coders dont need to think? 18:34 <+bridge> no 18:34 <+bridge> :( okay ill ask somewhere else 18:34 <+bridge> its not safe at all 18:34 <+bridge> https://tenor.com/view/mr-krabs-money-spongebob-gif-16807794896076753543 18:34 <+bridge> gg then 18:34 <+bridge> code it in lean 18:34 <+bridge> so ucan formallyp roof ddnet 18:34 <+bridge> works 18:34 <+bridge> its the truest safe 18:34 <+bridge> haskell boys 18:34 <+bridge> @ryozuki do some infighting, he hates Rust you know 18:34 <+bridge> https://lean-lang.org/ 18:34 <+bridge> lean sux 18:34 <+bridge> idc anymore humans will stop coding 18:34 <+bridge> @grok is this true 18:35 <+bridge> im a doomer 18:35 <+bridge> Hey, jugaras springlobe 4 el sabado? 18:35 <+bridge> snap builder C99 remake??!? 18:35 <+bridge> xD next? maybe 18:35 <+bridge> why 18:35 <+bridge> we gonna make a spanish team :D 18:35 <+bridge> i will keep coding just for you 🙂 18:35 <+bridge> human coding will be named "apecoding" 18:35 <+bridge> Still haven’t really had my wow moment with AI. It just makes so many stupid mistakes still, as soon as you get past the basics 18:36 <+bridge> yea 18:36 <+bridge> fun read https://rsaksida.com/blog/ape-coding/ 18:36 <+bridge> <_laveer> cuz claude code is being rewritten in rust 18:36 <+bridge> <_laveer> all roads lead to rust 18:36 <+bridge> nah 18:36 <+bridge> mine dont 18:36 <+bridge> which ai u using 18:36 <+bridge> i was wondering how people actually use ai to build things i guess they are the goats at prompts 18:36 <+bridge> i saw how Teero formats decimals, dont let him near my baby! /s 18:36 <+bridge> its jsut a wrapper for llm anyway 18:37 <+bridge> typical %a hater 18:37 <+bridge> and ai is best with typescript and web langagues so thats what they use 18:37 <+bridge> wat 18:37 <+bridge> lowkey ive had more success with non webapps lol 18:37 <+bridge> is it not? 18:37 <+bridge> maybe cuz all the webapps there are already poisoned by gpt3 code or smth. idk 18:37 <+bridge> I tried ChatGPT a bit, but that’s obviously bad. Claude does better, but still I was working on writing a small exploit a week ago, it just kept making mistakes with endianness 18:38 <+bridge> all "remake minecraft with ai" etc videos do a web page 18:38 <+bridge> btw "claude code" is what u call a harness, its not just a ai wrapper, it has systemp prompts and tools, etc 18:38 <+bridge> yeah i think AI has its strength in python/js/ts/html/css etc 18:38 <+bridge> claude is way different than chatgpt, try it 18:38 <+bridge> I usually need multiple waves to get what i want out of a webapp, to the point where i should just edit the typescript myself for some changes 18:39 <+bridge> but it can 1shot a lot of changes i ask in the tw codebase for example 18:39 <+bridge> what about compared to codex 18:39 <+bridge> chatgpt is codex 18:39 <+bridge> i just asked gemini 3.1 preview to find issues in my code. it found 3 "bugs" that are all intentional code design by me and have nothing to do with the actual problem that im having 18:39 <+bridge> chatgpt is general purpose 18:39 <+bridge> ChatGPT is intentionally bad, its to brainwash the masses 18:39 <+bridge> xd 18:39 <+bridge> Finding bugs I have had 0 success with. It especially struggles with C 18:39 <+bridge> what i mean is codex is made by chatgpt company openai 18:39 <+bridge> claude is by anthropic 18:40 <+bridge> same 18:40 <+bridge> yeah but codex does stuff differently than the web chatgpt 18:40 <+bridge> yeah 18:40 <+bridge> Depends on how you interact with it aswell, not saying you did wrong 18:40 <+bridge> Some other people seem to find very high quality bugs with AI, so I must just be holding it wrong, but for C and C++ it keeps giving me dated advice, or flat out incorrect stuff about UB 18:40 <+bridge> do u tell it to reference the c standard, passing it a link to a pdf or smth 18:41 <+bridge> more data better outcome 18:41 <+bridge> need a good prompt 18:41 <+bridge> idk, most of the time its wrong it just hallucinates things and bases its bugs on that. for example adding random null checks even though the function is guaranteed to always be called with valid data 18:41 <+bridge> OH, I did something very interesting a couple weeks ago. I tried to get it to emit correct ror61 for a 80 bit integer (with a struct). It struggled greatly 18:42 <+bridge> btw llms are better at rust i think 18:42 <+bridge>