00:47 <+bridge_> bro 00:47 <+bridge_> https://cdn.discordapp.com/attachments/293493549758939136/1516937244546437150/1.jpg?ex=6a347566&is=6a3323e6&hm=faafd17823b0720346ba8921c14fa2aa21cdaa4cf9b0d6f9a1e0433e5f018a89& 00:47 <+bridge_> https://cdn.discordapp.com/attachments/293493549758939136/1516937244852752505/2.jpg?ex=6a347566&is=6a3323e6&hm=d1d37013ebe8702970c87439133e8ff472d192354e64080bd75212ca15fab7db& 00:47 <+bridge_> https://cdn.discordapp.com/attachments/293493549758939136/1516937245125378298/3.jpg?ex=6a347566&is=6a3323e6&hm=67f536805a551b58f613faef22f30fea137dc3a94e2651a1e67c2a8d8f2ef317& 00:47 <+bridge_> https://cdn.discordapp.com/attachments/293493549758939136/1516937245402075188/4.jpg?ex=6a347566&is=6a3323e6&hm=608cdba6c43ec9530da008ce118dc356443ba40537545423694d1c5ffd26f730& 02:31 <+bridge_> erm 02:31 <+bridge_> @soulyvevo ChillerDragon: rock paper scissors each other, person who loses has to try to host volleybot again 03:04 <+bridge_> <0xfaulty> 20000 not banned long time already, other reason I think, will check later 03:23 <+bridge_> just let tpl take over volley 04:04 <+bridge_> i cant even get this to work xd 04:10 <+bridge_> 😮‍💨 I'll set it up entirely if you give me the VPS it should run on. 04:28 <+bridge_> ? 04:37 <+bridge_> ? 07:50 <+bridge_> !remind 67 try voli boll deploy again 07:50 <+bridge_> Maybe melon can support me debugging lua 09:30 <+bridge_> what is that bot thing 👀? 09:38 <+bridge_> anyone an idea for a nice profile pic for me on github? Maybe I'll just end up with the discord one 09:40 <+bridge_> 1x1-black.png should work 09:54 <+bridge_> Atomausstieg 👀 09:56 <+bridge_> Essigautomat 09:59 <+bridge_> @jxsl13 woah the teetui is really impressive 09:59 <+bridge_> the visual mode works so well oO 10:01 <+bridge_> I think you'd use a table for that data normally, not a diagram 10:02 <+bridge_> :deen_star: btw, there is no extra meaning about it 10:02 <+bridge_> ah, other people also said that 10:02 <+bridge_> yes we switched to tables later, I updated the PRs, they now contain both 10:03 <+bridge_> https://github.com/AssassinTee/ddnet-benchmarks 10:03 <+bridge_> 10:03 <+bridge_> My benchmarking repo - it got a lot more complex over time, more then I anticipated 10:04 <+bridge_> link? 10:06 <+bridge_> I just noticed, can demos be shared under the MIT license? 10:07 <+bridge_> you'd usually use a CC-* license for data files like demos 10:07 <+bridge_> why? 10:07 <+bridge_> guess i got it wrong then :/ 10:07 <+bridge_> it's probably not wrong to share them under MIT 10:08 <+bridge_> @blaiszephyr @gorp_tw https://paste.zillyhuhn.com/OL 10:08 <+bridge_> dis ma lua error 10:08 <+bridge_> but it's a bit weird, because it seems to be a software license, not a data license 10:10 <+bridge_> `luajit: main.lua:17: module 'lua-dotenv' not found:` \*slap\* 10:12 <+bridge_> https://github.com/ddnet/ddnet-maps doesn't contain any license :justatest: 10:23 <+bridge_> ty :0 :deen_star: 10:23 <+bridge_> try playing around with the min fps settings, maybe that helps improve the feedback. 10:23 <+bridge_> and rendering 10:24 <+bridge_> double jump does't work sadly, cuz there are no key released events in the terminal. 10:29 <+bridge_> what is volleybot 👀? 10:35 <+bridge_> Stats bot 10:35 <+bridge_> With website 10:35 <+bridge_> I never used luajit maybe it doesn’t work with luarocks out of the box 10:36 <+bridge_> why is lua in there? 10:44 <+bridge_> @jxsl13: I would assume a lot of the rendering is also stuttering because of the missing dead reckoning 10:44 <+bridge_> Even tho your readme says antiping xd 10:44 <+bridge_> Just on H instead of h where I am used to it being 10:44 <+bridge_> But so random that it has my reply to ping bot in there and it works 10:52 <+bridge_> can you open an issue on the twclient (i guess) explaining the problem a little bit more in depth. i'm not too deep into the ddnet code. 10:52 <+bridge_> chillerdragon 10:57 <+bridge_> tje problem with "dead reckoning" 10:57 <+bridge_> tje problem with "dead reckoning" 10:57 <+bridge_> the* 11:02 <+bridge_> https://discord.gg/graphicsprogramming 11:02 <+bridge_> there are some really nice graphics over there, especially the sdfs. good for pfps unless you are looking for smth else 11:04 <+bridge_> This discord would be interesting for me, but for any other reason than profile pics xD 11:06 <+bridge_> @essigautomat *prove 11:06 <+bridge_> in ur townhall msg 11:06 <+bridge_> DMs do exist 😉 11:06 <+bridge_> mb 11:07 <+bridge_> we should add a spellchecker to all of my messages 11:07 <+bridge_> would a spellchecker catch that? 11:08 <+bridge_> a good one, that also checks for grammer, yes 11:08 <+bridge_> a good one, that also checks for grammea, yes 11:08 <+bridge_> a good one, that also checks for grammeaH, yes 11:08 <+bridge_> xddddd 11:11 <+bridge_> I heard LLMs are really good a context-aware spell check 11:11 <+bridge_> You all must wonder, how I am able to program in C++, while I am not able to write properly 11:12 <+bridge_> C++ compiles, if you name all your variables **consistenly** wrong 11:14 <+bridge_> i don't think it's possible to do it properly in online chats, peoples names may be normal words like melon, you would need to give it all the usernames that are often used or something similar. 11:14 <+bridge_> I am pretty sure, the AI can handle that for 99% 11:15 <+bridge_> If i say "Melon killed me yesterday." 11:15 <+bridge_> it would correct me to 11:15 <+bridge_> "A melon killed me yesterday." 11:17 <+bridge_> it would ask whether you are aghost 11:17 <+bridge_> you basically need to give it the ddnet wiki so it knows all the made up words that we use ingame and a list of people that chat regularly 11:17 <+bridge_> :kek: 11:18 <+bridge_> you can also jailbreak llm translators 11:18 <+bridge_> i think it was this article: https://www.coindesk.com/markets/2026/04/08/move-over-bitcoin-and-quantum-risks-anthropic-s-mythos-ai-changes-everything-for-defi?referrer=www.coinrithm.com%2Fen%2Fcryptocurrencies%2Fmythos 11:19 <+bridge_> and a similar case happened with antrophics new model `fable` - hence why its currently unavailable: https://www.anthropic.com/news/fable-mythos-access 11:19 <+bridge_> i guess it would be nice if the ddnet bot could translate simple messages using some lightweight model like qwen 0.8B 11:21 <+bridge_> i also had a good chuckle reading this: https://www.theregister.com/special-features/2024/10/02/ai-agent-promotes-itself-to-sysadmin-breaks-boot-sequence/1183461 :kek: 11:27 <+bridge_> the blog post referenced only talks about "Mythos Preview identified a number of weaknesses […]" 11:28 <+bridge_> not "system-critical vulnerabilities" 11:28 <+bridge_> the blog post referenced only talks about "Mythos Preview identified a number of weaknesses […] in SSH […]" 11:32 <+bridge_> Seems like i conflated weakness with vulnerability - my bad 11:41 <+bridge_> I mean a completely random human would also be lost. If I asked my english teacher whether the sentence "Melon killed me yesterday" is correct she'd have trouble too. What's nice is that unlike traditional autocorrect you can at least feed the context into this one and expect to get pretty good results 11:46 <+bridge_> bru i was about to touch code i have unit tests for just to realize i cant run unit tests anymore -.- 11:46 <+bridge_> every fakin time 11:46 <+bridge_> sexmentation faulting in score\_test.cpp `SingleScore_TopRegional_Test::AddToRegistry` even when filtering out that test 11:46 <+bridge_> ima open angry github issue and complain to these ddnet devs 11:46 <+bridge_> oh no this time not even build dir wipe or reboot helped omagawd what a side quest 11:48 <+bridge_> im just going to throw a guess that somehow your code links with globally installed gtest 11:50 <+bridge_> @chillerdragon I can not reproduce this 11:51 <+bridge_> oh I can reproduce this 11:51 <+bridge_> im just going to throw a guess that somehow your code links with globally installed gtest, and since `-DCMAKE_BUILD_TYPE=Debug` builds with debug checks, there's ABI mismatch 11:52 <+bridge_> I am starting the debugger rn 11:53 <+bridge_> looks like a bug in gtest to me 11:58 <+bridge_> Hmm, am I reading `CMakeLists.txt` correctly, it doesn't download GTest if there's globally installed one? 11:58 <+bridge_> even if user passes `-DDOWNLOAD_GTEST=ON` 12:00 <+CN-Bridge> yeah, it would use the system GTest if there is one. 12:00 <+bridge_> that's not good 😬 12:01 <+CN-Bridge> why? 12:03 <+bridge_> I am checking if we need INSTANTIATE_TEST_SUITE_P here 12:03 <+bridge_> https://github.com/ddnet/ddnet/blob/81fddacce66696d8fd2d07c4aa21d87943e30cfa/CMakeLists.txt#L3943-L3944 because of these 2 lines 12:05 <+bridge_> when C++ is compiled with these definitions, it's not compatible with regular C++ 12:05 <+bridge_> when C++ is compiled with these definitions, it's not compatible with C++ compiled without them 12:14 <+bridge_> W 12:14 <+bridge_> @milkeeycat goated guess, I came to the same conclusion 12:16 <+bridge_> gtest is required by abseil-cpp which is required by lots of dependencies 12:17 <+bridge_> kinda sad that a package like this _requires_ gtest 12:26 <+bridge> @chillerdragon do you think it makes sense to check for the debug option in order to download gtest? 12:27 <+bridge> Even writing this down makes me think no 12:47 <+bridge> @learath2 about https://github.com/ddnet/ddnet/issues/8410: what's our current stance on how to implement it? 12:47 <+bridge> custom protocol or use something like etcd? 12:48 <+bridge> what does it need to do? 12:48 <+bridge> - synchronize bans across game servers in a global list 12:48 <+bridge> - allow executing commands on all running servers 12:48 <+bridge> - allow changing config options globally 12:50 <+bridge> for the bans, it needs to store a list and be able to send updates. it should ideally not break and re-synchronize when we lose connection. local bans should be possible as well, without interfering with global ones 12:51 <+bridge> it'd be good if each location was able to start servers with a last-known version of the bans, so that global connection is not necessary to get (stale) bans or configuration 12:52 <+bridge> (I'll put our discussion into the issue after it concludes) 12:53 <+bridge> Bans also have a performance requirement that the queries are as fast as possible, so if the solution we come up with is external, we might need to implement an in memory cache (could be shared accross all gameservers on the node) 12:54 <+bridge> Every location probably needs to replicate whatever the list with an eventual consistency promise. I think this is also sane for the config 12:54 <+bridge> yes, the idea of that external service/own protocol would be getting updates and including those in the game server's own data structures 12:55 <+bridge> yes, I'd really like eventual consistency over stronger consistency, unfortunately I haven't found an eventual-consisten etcd 12:55 <+bridge> yes, I'd really like eventual consistency over stronger consistency, unfortunately I haven't found an eventual-consistent etcd 12:56 <+bridge> I have been thinking about this a little and I think maybe we should just limit our scope to bans 12:57 <+bridge> we can do that at first 🙂 12:57 <+bridge> The problem on the config side is not really that we don't have systems to change configs or execute commands. The problem there is monitoring, we have no idea what server is running what configuration 12:57 <+bridge> We have no idea if our commands succeed 12:57 <+bridge> but we control the game servers as well 12:57 <+bridge> we can make them listen to such a service, and respond which command executions worked 12:57 <+bridge> etc. 12:58 <+bridge> but yea, limiting it to bans first sounds like a sane strategy 12:58 <+bridge> I thought that aswell initially, but thinking about it later, I kinda didn't see the point of having econ, rcon, fifo + now one more protocol 12:59 <+bridge> econ, rcon, fifo are all just interfaces to a command line 12:59 <+bridge> I think it'd make sense to have more structured access over a new protoocl 13:00 <+bridge> the current global bans system is very bolted n 13:00 <+bridge> the current global bans system is very bolted on 13:00 <+bridge> it can't really keep track of which bans are applied and it spams the log output when initially starting, and it doesn't understand ban expiry 13:05 <+bridge> I mean very initially I had this whole idea of a "service"/"events"/"bus" architecture, where all the "nodes" would advertise services: 13:05 <+bridge> `ddnet.org` would advertise the service `bans_global[@infra.ddnet.org]` 13:05 <+bridge> `ger10.ddnet.org` would advertise `bans_cache` 13:05 <+bridge> `ger10.ddnet.org:8303` would advertise stuff like `events, chat, player_info, config...` 13:05 <+bridge> 13:05 <+bridge> Then each node could request any service from any other node. Which would allow stuff like a moderator control panel node that can just subscribe to the `chat` stream from a server to give live chat, or a watchdog node subscribe to the `events` service to monitor all critical events across all servers 13:05 <+bridge> yes, that part is objectively disgusting, that's why I thought maybe THAT specific thing needs to become custom, not other stuff 13:06 <+bridge> ah, I wouldn't try to engineer it that much 13:06 <+bridge> my custom protocol would maybe be like this: 13:06 <+bridge> Something like NATS? that's a decent chunk to implement properly 13:07 <+bridge> client sends "subscribe bans" — server replies with current list of bans — server sends updates to bans as they come in 13:07 <+bridge> Perhaps? I've never heard of it before, but it is basically a message queue with a router 13:08 <+bridge> basically only three messages, but they cover everything we need. we extend it by extending protocol 13:09 <+bridge> would need a decision on the shape of the protocol and the underlying transport 13:09 <+bridge> simplest would be something like JSON lines over plain TCP 13:10 <+bridge> I don't think there is any point to extending that is the conclusion I've come to. Adding any extra useful functionality on top of that requires so much change that we will never do it 13:11 <+bridge> what context would you give it? the last x messages? i could say melon killed me completely unrelated to a conversation 13:11 <+bridge> Like it can be shoehorned into delivering config changes or commands too, but we haven't solved anything there with that, delivering config changes over ssh/fifo is no worse than that. We still don't get proper monitoring of the running config or the outputs of our commands 13:13 <+bridge> Well I'd give it a context blob depending on well the context. For discord a list of all members names might be good, all channel names, guild information, perhaps a tiny prompt saying that discord uses a flavor of markdown and some symbols are expected and are not grammatical issues. 13:14 <+bridge> Previous x messages is also probably a good addition 13:17 <+bridge> Just capturing what the server already knows would likely solve the bulk of this without any upstream changes. All we need then is a source of truth to keep everything in sync 13:17 <+bridge> (btw the example you gave was kinda a grammatical error, so technically spellcheck-wise it has no issues 😄 ) 13:18 <+bridge> Yes that is correct and indeed a bit odd. I am used to passing it because otherwise it used to say no such target run\_tests for me. Maybe making it prefer the downloaded one if explicitly asked for is already an improvement 13:18 <+bridge> I don’t think this is possible. Check debug where? We can check if ddnet is being built with debug but as far as I can tell we can’t check if the system installed gtest is built with this debug option 13:18 <+bridge> But even if it was optional I think it’s reasonable to support users that have a popular library installed on their system. 13:18 <+bridge> Yea that’s troll should be a optional install dependency and being uninstalled after 13:19 <+bridge> the convo originated from melon making a grammatical error in a town hall post, so he suggested a spellcheck, but the spellcheck wouldn't have helped, so talked about using llms 13:19 <+bridge> the convo originated from melon making a grammatical error in a town hall post, so he suggested a spellcheck, but the spellcheck wouldn't have helped, so we talked about using llms 13:19 <+bridge> please someone fact check whether an LLM proposes to add an article 13:20 <+bridge> Please spell-check my message to be sent to the DDNet discord: "Melon killed me". Only send the correction, or the original if there are no errors. 13:20 <+bridge> (inb4 usage policy) 13:22 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517127238427086979/image.png?ex=6a352658&is=6a33d4d8&hm=42d9f371e2b856e65074a3ce3021f44d2706e70d9f869c06189a6c29ca652b48& 13:22 <+bridge> > You are a grammar/spellcheck AI, I will give you a string and you will reply with the fixed string, nothing else. 13:22 <+bridge> starting with: "Melon killed me yesterday in a far run of back in time 2" 13:22 <+bridge> 13:22 <+bridge> response: 13:22 <+bridge> 13:22 <+bridge> > Melon killed me yesterday in a far run of Back in Time 2. 13:22 <+bridge> sounds good 🙂 13:22 <+bridge> > You are a grammar/spellcheck AI, I will give you a string and you will reply with the fixed string, nothing else. 13:22 <+bridge> > starting with: "Melon killed me yesterday in a far run of back in time 2" 13:22 <+bridge> 13:22 <+bridge> response: 13:22 <+bridge> 13:22 <+bridge> > Melon killed me yesterday in a far run of Back in Time 2. 13:22 <+bridge> Pro AI user 13:23 <+bridge> so it seems like a hypothetical problem and LLMs are indeed quite good spell checkers 😉 13:23 <+bridge> xd i guess so 13:23 <+bridge> I'm sure there is stuff they get wrong 13:23 <+bridge> they're famously right most of the time, and confidently wrong otherwise 13:23 <+bridge> @murpi time to add llm to the ddnet bot and run every single message through it so we all have perfect grammar 13:23 <+bridge> no pls 13:23 <+bridge> I’m 100% sure we can find a pathological case. But yeah they should be pretty good, they are afterall kinda language experts 13:24 <+bridge> no pls 13:25 <+bridge> interesting CPU bug: https://trifectatech.org/blog/zlib-rs-in-firefox/ 13:25 <+bridge> Cpu bug? Now that’s one you don’t hear often 13:27 <+bridge> Here is one normal spellcheck can't possibly do 13:27 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517128672996491436/image.png?ex=6a3527ae&is=6a33d62e&hm=db85475f63c0c6fb63579a0c202576ea4dd378a0a6047578b0c6ca53847126a1& 13:28 <+bridge> I'd guess normal spell check in MS word or similar works 13:29 <+bridge> reminds me of https://gmplib.org/ destroying ryzen 9000s 13:29 <+bridge> You kinda need NLP for this, no? Which "then" you are supposed to use requires understanding the sentence structure 13:30 <+bridge> yes, I'd assume MS word is doing some NLP 13:32 <+bridge> try making a sentence with @alw5 13:33 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517130013726740531/image.png?ex=6a3528ee&is=6a33d76e&hm=44de884e50d97f910d5eba9a27eb28c96153eb46a3d8b7b2e151e960b2d03bc8& 13:33 <+bridge> drop the @ xd 13:34 <+bridge> AHA, ladies and gentleman, we gottem 13:34 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517130394548568256/image.png?ex=6a352948&is=6a33d7c8&hm=e290dd554f654fc9a7cb7809b195f2b9565c582eedfcd4141a67c89e6492cfaa& 13:35 <+bridge> I mean the comma is good 13:35 <+bridge> Hm, actually yeah that does sound okay 13:35 <+bridge> but yea, capitalization is opinion 13:36 <+bridge> I prefer not capitalizing heinrich5991 at the start of the sentence, but 🤷‍♀️ 13:36 <+bridge> Heinrich5889 13:36 <+bridge> my local Qwen3.7 27B said this: 13:36 <+bridge> ```json 13:36 <+bridge> { 13:36 <+bridge> "errors": [ 13:36 <+bridge> { 13:36 <+bridge> "start": 12, 13:36 <+bridge> "end": 25, 13:36 <+bridge> "suggestion": "always" 13:36 <+bridge> } 13:36 <+bridge> ], 13:36 <+bridge> "formatted": "I find that always always finds something to talk about" 13:36 <+bridge> ``` 13:36 <+bridge> The issue: "always always" is a duplicated word — likely just needs to collapse to a single "always." 13:36 <+bridge> ``` 13:36 <+bridge> You know what, it does seem to understand that this "always" is used as a name and capitalizes it within sentences too, but I think that's just my context leaking 13:36 <+bridge> my local Qwen3.7 27B said this: 13:36 <+bridge> ```json 13:36 <+bridge> { 13:36 <+bridge> "errors": [ 13:37 <+bridge> { 13:37 <+bridge> "start": 12, 13:37 <+bridge> "end": 25, 13:37 <+bridge> "suggestion": "always" 13:37 <+bridge> } 13:37 <+bridge> ], 13:37 <+bridge> "formatted": "I find that always always finds something to talk about" 13:37 <+bridge> ``` 13:37 <+bridge> The issue: "always always" is a duplicated word — likely just needs to collapse to a single "always." 13:37 <+bridge> yea, I wanted to ask you to try without the context 😄 13:37 <+bridge> but then I apparently forgot 13:37 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517131005503344762/image.png?ex=6a3529da&is=6a33d85a&hm=fe14485012dcb9571f8b7854c6cdef6149a633e81dde89fc888264f0f0af873a& 13:37 <+bridge> can you try the 0.8B? the ddnet bot def cant run the big ones 13:37 <+bridge> No no, it suggest to add another "always". 13:37 <+bridge> uh let me check if i can quickly grab it 13:37 <+bridge> I can't really, because openai ruined chatgpt by making it leak across chats. If you ask chatgpt that it'll vehemently deny it 13:37 <+bridge> great 13:38 <+bridge> lol wont that make individual chats a lot worse? 13:38 <+bridge> i run small models locally, my GPU is perfectly capable of holding enough context for my needs :D 13:38 <+bridge> well, "small" - 8b qwen 3.6 is not really small :D 13:39 <+bridge> well, "small" - Q8 27B qwen 3.6 is not really small :D 13:39 <+bridge> It makes it so very annoying not being able to get a clean slate. They don't seem to leak the entire thing but they kinda seem to run another model over your chat that creates a summary and extracts interesting unique points from it, then that summary gets fed into it 13:40 <+bridge> Like 2 days ago I asked it to translate and summarize a house listing, it did that part fine, then it proceeded to tell me how my income as a student wouldn't be enough to afford it. I checked it's not in its long term memory that I'm a student 13:40 <+bridge> :ToadStunned: 13:40 <+bridge> Like 2 days ago I asked it to translate and summarize a house listing, it did that part fine, then it proceeded to tell me how my income as a student wouldn't be enough to afford it. I checked, it's not in its long term memory that I'm a student 13:41 <+bridge> oops 13:41 <+bridge> the heuristic is 1/3rd of your income for rent - which is in parts of the world plain unrealistic 13:42 <+bridge> In amsterdam a lot of landlords want you to have 4x the income, which combined with amsterdam rents means that I'd need to be earning something like 10k to be able to rent a random ass apartment 😄 13:42 <+bridge> amsterdam mentioned 13:42 <+bridge> pog 13:43 <+bridge> hi chiller 13:43 <+bridge> @teero777 13:43 <+bridge> ```json 13:43 <+bridge> { 13:43 <+bridge> "errors": [ 13:43 <+bridge> { 13:43 <+bridge> "for_mof_start": 0, 13:43 <+bridge> "end_location": 23, 13:43 <+bridge> "suggestion": "find", 13:43 <+bridge> "formatted": "I find that always **always** finds something to talk about" 13:43 <+bridge> } 13:43 <+bridge> ] 13:43 <+bridge> } 13:43 <+bridge> ``` 13:43 <+bridge> elo tero 13:43 <+bridge> They are just basically telling people to just leave this city unless they are an executive or senior senior senior 13:43 <+bridge> anyway, very off-topic 13:44 <+bridge> thats qwen3.5-0.8b 13:44 <+bridge> @teero777 13:44 <+bridge> ```json 13:44 <+bridge> { 13:44 <+bridge> "errors": [ 13:44 <+bridge> { 13:44 <+bridge> "for_mof_start": 0, 13:44 <+bridge> "end_location": 23, 13:44 <+bridge> "suggestion": "find", 13:44 <+bridge> "formatted": "I find that always **always** finds something to talk about" 13:44 <+bridge> } 13:44 <+bridge> ] 13:44 <+bridge> } 13:44 <+bridge> ``` 13:44 <+bridge> 13:44 <+bridge> thats qwen3.5-0.8b 13:44 <+bridge> nice, without thinking? 13:44 <+bridge> do /nothink 13:44 <+bridge> ChatGPT: 13:44 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517132941195612180/image.png?ex=6a352ba8&is=6a33da28&hm=aba96ec43448d2fc9ac49b89add3f2879ff157385681bd95449667540ed1e05e& 13:44 <+bridge> The metadata is all wrong lol 13:44 <+bridge> yes, 0.05s - 73 tokens 13:46 <+bridge> i wish i could do that without thinking 13:46 <+bridge> I saw this example in a video where they were benchmarking multiple AIs, some _did_ manage to get this right 13:46 <+bridge> One of the AIs asked you, if you want to go to the car wash in order to clean your car or to play some bills/visit some coworkers 13:46 <+bridge> ChatGPT 5.5 doesn't get this correct even with high thinking btw 13:46 <+bridge> reminds me of the strawberry test xd 13:47 <+bridge> well technically its correct, what if you work there, at some car washes you can also buy stuff 13:48 <+bridge> reminds me of this 13:48 <+bridge> ``` 13:48 <+bridge> 13:48 <+bridge> "there are 2 r's in strawberry" 13:48 <+bridge> "st >r< a w b e >r< >r< y" - no wait.. thats 3.. wait 13:48 <+bridge> -- thinking for 39928 seconds 13:48 <+bridge> "there is 1 r in stawbery" 13:48 <+bridge> ``` 13:48 <+bridge> Even with this prompt saying I want to get my car washed... 13:48 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517133808665628883/image.png?ex=6a352c76&is=6a33daf6&hm=323161cd8feff4a0c9d509bf0ce9d8c6ff53bfd54c3c4c13beaff670ad18b9ce& 13:48 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517133898478391299/image.png?ex=6a352c8c&is=6a33db0c&hm=5c451dbe8373bc5167a1b69a7fe3394c6602d7c3ff182cb4443dfcb077f5096c& 13:48 <+bridge> Just did that too :kek: 13:48 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517133983610048564/image.png?ex=6a352ca0&is=6a33db20&hm=58202193bb500eb2120733a9380702a122869ebf5061987235471aa88aadf266& 13:49 <+bridge> Yeah AI will replace us, it definitely has thinking capabilities. AGI tomorrow bro 13:49 <+bridge> <01000111g> such a joke. Living should be affordable, but modern democracies dont even get that one right, maybe with Vienna as an exception 13:49 <+bridge> Fable? Oh wait 13:49 <+bridge> i don't have any Internet wtf 13:49 <+bridge> Vienna is such an outlier it's insane 13:50 <+bridge> Vienna is such an outlier, it's insane 13:50 <+bridge> lmao 13:50 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517134296056594522/Bildschirmfoto_20260618_134950.png?ex=6a352ceb&is=6a33db6b&hm=6b9de2bc60c264de273da2d96a544ba290c88558bc12ed973d2d05f2b981a09c& 13:50 <+bridge> US says nurr 13:50 <+bridge> `Plan - Big Pickle` 13:50 <+bridge> Ok this one can replace me, I'll allow it 13:50 <+bridge> I can't fkn send anything 13:50 <+bridge> send help 13:50 <+bridge> I bet grok gets it too btw, I find it is usually smarter 13:50 <+bridge> I bet grok gets it too btw, I find it is usually "smarter" 13:50 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517134511266205817/HKdAH-UbgAEQK5M.png?ex=6a352d1e&is=6a33db9e&hm=0e8dae22b7f81ddf0dcf30f7448631f60144c226fb959c14efe8345d09f7e073& 13:51 <+bridge> wow now it sent it 13:51 <+bridge> after the 5th time 13:51 <+bridge> lets see if it sends the other 4 ones too 13:51 <+bridge> I've seen a lot of those on reddit :kek: 13:51 <+bridge> If only I was a US citizen I could have access to god 13:52 <+bridge> Oh btw did you hear about the part where they leaked Mythos weights? 13:52 <+bridge> cool 🪤 13:52 <+bridge> 700T parameters supposedly :nouis: 13:53 <+bridge> https://www.youtube.com/shorts/ASsK8fCg6Cw 13:53 <+bridge> btw hearing export restrictions on software sent me right back to 1990s 13:53 <+bridge> Well we did have that already... for some military use software 13:54 <+bridge> i really hope some employee at anthropic decides to not gaf and leak the entire thing 13:55 <+bridge> the chinese would have fun with it 13:55 <+bridge> given how much AI engineers usually make anually, i doubt it :D 13:57 <+bridge> I have a profound distaste with how these companies just stole everything remotely public without giving a fuck about licenses and IP, then are now not only gatekeeping the fruits of their thievery but are also profiting from it 13:57 <+bridge> Disgusting 13:57 <+bridge> how else do you get that much info 13:57 <+bridge> I'd have done the same thing 13:58 <+bridge> Given how harmful these companies are a lot of their employees feel quite conflicted about working there 13:58 <+bridge> I can tell from the only person I ever met working at microslop 13:58 <+bridge> @chillerdragon this bot bugged? 13:58 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517136424216952903/image.png?ex=6a352ee6&is=6a33dd66&hm=46b8f42e841b2d5ce81ec8ad445986f8f0cc05d017048f74f8f3af78d344c4b1& 13:58 <+bridge> its spamming my stats 13:58 <+bridge> holy spam 13:59 <+bridge> ChillerDragon .... 13:59 <+bridge> ban chillerdragon, right now 14:03 <+bridge> https://cdn.discordapp.com/attachments/293493549758939136/1517137694537285713/image-56.png?ex=6a353015&is=6a33de95&hm=8bb4ea9b878b86214b982bf2988499acf928b273648cf351ce81055b7d580b37& 14:08 <+bridge> chiller supremacy 14:08 <+bridge> <01000111g> yeah it is crazy, at this point i wish it did not happen at all. I could just chill and scroll a bit of stackoverflow to build something in a month, and not have everyone hyperactively publishing "prod-ready" bloat. And I'm getting baited myself as well. 14:09 <+bridge> I would immediately cancel my chatgpt sub if it wasn't a requirement for being hireable in the big 26 14:10 <+bridge> Literally have to force myself to slop it up so I can maybe get a job 14:10 <+bridge> <01000111g> I now witness students outsourcing their master's thesis to claude and they literally cant answer a simple scientific question. Its like proxy prompting. You prompt the student and he prompts his agent???? xd comedy gold 14:11 <+bridge> Did you see that Bun PR/issue/PR chain where it was basically Claude-1, Claude-2, Claude-3 and Claude-4 iterating on eachothers work by prompting eachother? 14:12 <+bridge> (just to come up with a shit solution that makes the test pass) 14:15 <+bridge> i hope 14:15 <+bridge> @teero777 did u try glm-5.2? 14:15 <+bridge> <01000111g> No, but sounds fun. I mean agentic coding is entertaining, but crazy bloat and token usage. And for simple things in ddnet client it works out of the box, which is crazy to me ( I experimented recently) 14:15 <+bridge> its open source 14:15 <+bridge> i think the gap is closing 14:15 <+bridge> and unironically, it will be thanks to the chinese 14:15 <+bridge> US is more restrictive than china rn 14:15 <+bridge> I need some help, ive been using a 60% keyboard for a while now but i just now started to use emacs with its default keybinds. usually i just have the arrow keys rebound to M-wasd but those binds are now necessary for using emacs so i cant use them anymore. M-hjkl also doesnt work since it conflicts. any idea what key combo i can use to get arrow keys? xd 14:16 <+bridge> not yey 14:16 <+bridge> not yet 14:16 <+bridge> programmers are the only ones who automate themselves out and cant stop it 14:16 <+bridge> <01000111g> But imagine some rich guy prompting 100 agents to build a ddnet account system. Will it ever work and be secure? or will noone get the codebase and possible exploits? do you need to pentedt with 100 agents as well. :justatest: 14:16 <+bridge> i read something about in aviation the same thing kinda happened 14:16 <+bridge> most planes are automatic 14:17 <+bridge> https://openlm.ai/glm-5.2/ ??? its better than gemini, at least on these graphs. i didnt expect that 14:17 <+bridge> glm 5.2 is the SOTA of open source 14:17 <+bridge> u cant run it locally tho its 750B 14:17 <+bridge> i use ollama cloud sub 14:18 <+bridge> i use it with open code and its not bad, i made some ffxiv plugins that worked nicely 14:18 <+bridge> At least in aviation it's not statistical black magic taking over, it's pretty much deterministic, so deterministic that almost every flight automation system has 2 backups that all need to agree on flight control decisions 14:18 <+bridge> ¯\_(ツ)_/¯ 14:18 <+bridge> Meanwhile my gpt is hallucinating the existence of random Intel Pin functions 14:18 <+bridge> unrelated but, it is "deterministic" for the same seed and input, just they put a random seed to make it more realistic/different 14:18 <+bridge> i have big swap dw /s 14:19 <+bridge> and the emergent behaviour is smth interesting 14:19 <+bridge> but yeah 14:19 <+bridge> sucks for humanity 14:19 <+bridge> just find the seed where it isn't hallucinating duh 14:19 <+bridge> 14:20 <+bridge> i would say "unpredicatble" 14:20 <+bridge>