00:41 < bridge> <djo_biden_> !help
03:33 < bridge> <swarfey> Discord had some issue with germans
03:33 < bridge> <swarfey> A ton of germans got flagged and accounts got disabled, including mine
03:34 < bridge> <swarfey> https://www.reddit.com/r/discordapp/s/hkNuyeLil1
03:34 < bridge> <swarfey> :kek:
04:20 < bridge> <triki> how to force vote in f2? i dont undestend syntax
04:56 < bridge> <chillerdragon> Common discord L
07:06 < bridge> <teero777> `vote yes` or `vote no`
07:54 < bridge> <blaiszephyr> @tsfreddie are we still sure we'd want to use a player flag for something that might as well be a netmsg ? Or is there a reason as to why this won't work
07:54 < bridge> <tsfreddie> ask heinrich actually
07:55 < bridge> <tsfreddie> i don't know either
07:55 < bridge> <tsfreddie> :justatest:
07:55 < bridge> <blaiszephyr> Didn't hear from Heinrich in a while :NotLikeKogasa:
07:55 < bridge> <tsfreddie> https://github.com/ddnet/ddnet/pull/9357#issuecomment-2546832457
07:56 < bridge> <blaiszephyr> oh! I see
07:56 < bridge> <tsfreddie> the compatibility thing i get, but i think adding more playerflags might be as easy as unclamp the protocol, it was not a byte, it is just varint but clamped as far as i can understand.
07:56 < bridge> <tsfreddie> haven't tried it tho
07:57 < bridge> <tsfreddie> i mean i can try it right now one sec
08:01 < bridge> <tsfreddie> seems fine to me
08:01 < bridge> <tsfreddie> https://cdn.discordapp.com/attachments/293493549758939136/1319560321437798454/image.png?ex=676667d0&is=67651650&hm=08621cca61099ff3df87ef2235b1783c4cf684cd772ad4abfc88b48edf4876a6&
08:02 < bridge> <tsfreddie> a version check to make sure you don't send out of bound flags to server with lower version and we are good to go
09:07 < bridge> <Assa> my account got deactivated for "suspicous activity" and I got a mail that "I got hacked" from discord. 2FA for the win
09:07 < bridge> <Assa> also I know this has to be bullshit
09:36 < bridge> <chairn> maybe discord got hacked and they won't acknowledge it
09:38 < bridge> <Assa> no company does, but they have to by law. But first they need to collect info about what was hacked and by what degree
09:39 < bridge> <Assa> this can range from the "secrataries ipad" to "all of our infrastructure can be thrown away"
09:53 < ws-client> <ChillerDragon> https://github.com/ddnet/ddnet/blob/b52f1a41a5890fca6da3a877221f7dce69eec60c/src/engine/shared/network.cpp#L361
09:54 < ws-client> <ChillerDragon> does someone know why we shift by 2 and not 4 here?
09:54 < ws-client> <ChillerDragon> according to the libtw2 docs a vital chunk header consists for 3 bytes and the sequence number takes the full third byte and half (4 bits) of the second byte
09:54 < ws-client> <ChillerDragon> https://github.com/heinrich5991/libtw2/blob/master/doc/packet.md
09:58 < ws-client> <ChillerDragon> the mask is a bit hard to read in ddnet ``pData[1] |= (m_Sequence >> 2) & (~((1 << Split) - 1));``
09:58 < ws-client> <ChillerDragon> in teeworlds 0.6 its just ``pData[1] |= (m_Sequence>>2)&0xf0;`` and ``0xf0`` in binary is ``0b11110000`` so you can clearly see the 4 bit mask
09:59 < ws-client> <ChillerDragon> so shouldn't it also be a 4 bit shift?
10:00 < ws-client> <ChillerDragon> imo sequence numbers up to 255 should only be written to the last byte and only 256 and higher should start leaking into the 2nd byte but from what i can tell the sequence number 64 already starts writing into the second byte
10:14 < ws-client> <ChillerDragon> okay looking at some traffic my ddnet client sent during a 0.6 connection it seems like it actually does not pollute the 2nd byte before it reaches 256 but how
10:14 < ws-client> <ChillerDragon> https://zillyhuhn.com/cs/.1734685973.png
10:17 < ws-client> <ChillerDragon> @heinrich5991 also in the dissector it says ``00.. .... 1111 1111`` for the sequence number. Shouldn't it be ``0000 .... 1111 1111`` ? First 4 then 8?
10:17 < ws-client> <ChillerDragon> https://zillyhuhn.com/cs/.1734686251.png
10:25 < ws-client> <ChillerDragon> oh no that makes no sense the size is not decimal 49 because the second byte is 0x31 the size also leaked into the first byte because it is bigger than 4 bits (15)
10:26 < ws-client> <ChillerDragon> the only flag bit that is set is vital so the first byte of the header should be 0x40 but it is 0x43 (0b01000011) so the last 2 bits are size okay now i understand nothing anymore
10:30 < ws-client> <ChillerDragon> that shit is more encryption than compression ngl
11:22 < bridge> <Assa> the line of this can be extremely blurry, I know a nintendo puzzle where they encrypted some data with bit operations and your task was to write a decoder
11:25 < bridge> <inv41idu53rn4m3> More encoding than encryption, after all I'm not seeing very many large primes or elliptic curves
12:58 < bridge> <zhn> quantum computer doesn't care about both :troll:
14:17 < bridge> <Jupstar ✪> what do you need this for?
14:18 < bridge> <Jupstar ✪> you can always printf whatever you want, but the fact that you don't know this and directly want it for a projectile is a bit sus to me
14:22 < bridge> <exosphere___endosphere> oh i just found this file, my original goal was just to study the laser code, i thought maybe something cool could be made out of it, but before that i tried to change the HUD, it's pretty cool to do that
14:22 < bridge> <exosphere___endosphere> There were no problems with the HUD, but with the laser I had my first problems understanding the code
14:22 < bridge> <Jupstar ✪> laser.cpp is really mostly the laser physics
14:22 < bridge> <Jupstar ✪> if you want to change the rendering, then the file is called items.cpp or smth like that
14:23 < bridge> <exosphere___endosphere> Why change the rendering? Will something change?
14:23 < bridge> <Jupstar ✪> I mean what part of the laser do you want to change?
14:23 < bridge> <zhn> what do you want in first place?
14:24 < bridge> <Jupstar ✪> Generally speaking you probably won't have an easy time to understand the laser in full detail.
14:24 < bridge> <Jupstar ✪> 
14:24 < bridge> <Jupstar ✪> The DoBounce method is the most interesting one
14:24 < bridge> <exosphere___endosphere> just output information about the laser to understand the code by type of direction, energy, bounces, etc.
14:25 < bridge> <Jupstar ✪> then simlpy print what you need in the tick function
14:25 < bridge> <exosphere___endosphere> I tried to understand through chat gpt, and I was very surprised with the laser's work)
14:25 < bridge> <Jupstar ✪> but if you already know about direction, energy and bounces. the variables are most likely also called similary to this
14:26 < bridge> <Jupstar ✪> yeah lasers are bit hacky, because they need to stay alive while the client renders them, so they don't directly despawn when they are gone/finished.
14:26 < bridge> <Jupstar ✪> 
14:26 < bridge> <Jupstar ✪> and if you look in ddrace code, you see all the 2 trillion extra cases for laser doors, tele lasers etc
14:27 < bridge> <Jupstar ✪> which makes it even harder to understand the whole code
14:27 < bridge> <Jupstar ✪> maybe first try:
14:27 < bridge> <Jupstar ✪> https://github.com/teeworlds/teeworlds/blob/c56fa9e6a20cfc9d7d16502e18c7d7633acdf492/src/game/server/entities/laser.cpp
14:29 < bridge> <Jupstar ✪> doors are extra code btw, so you gladly don't have to deal with that 😂
14:42 < bridge> <zhn> huhu
14:42 < bridge> <Jupstar ✪> hai
14:42 < bridge> <_voxeldoesart> ho ho ho
14:42 < bridge> <Jupstar ✪> zanta
14:46 < bridge> <zhn> zaza
17:27 < bridge> <sannitt.> hi
17:45 < bridge> <blaiszephyr> @tsfreddie can you give me the static clang format 10 link you mentioned heinrich sent? i can't find it and i want to add it to the readme, new contributors often forget
17:46 < bridge> <_qey> I though he was Chinese
17:46 < bridge> <blaiszephyr> and? :kek:
17:46 < bridge> <blaiszephyr> he a silly billy
17:46 < bridge> <_qey> Chinese from Europe
17:47 < bridge> <_qey> Maybe we shouldn’t combat blockers on non-blocker servers, maybe we should create more blocker maps? I’m tired of banning those idiots
18:03 < bridge> <blaiszephyr> @askll_star this could be helpful for you :P
18:03 < bridge> <blaiszephyr> 
18:03 < bridge> <blaiszephyr> our fix_style script is a bully
18:10 < bridge> <askll_star> @ml1xx I've tried. It doesn't give me any warn.
19:42 < bridge> <htrudfhdfhuytryfghnj> Now it's clear, This message is visible only to you (yourself) I'll change my nickname okay? plz dont ban me...
23:08 < bridge> <ryozuki> https://en.wikipedia.org/wiki/Brussels_effect
23:17 < bridge> <totar> New openai model has 2700 code forces rating... it's so over
23:17 < bridge> <totar> https://cdn.discordapp.com/attachments/293493549758939136/1319790862951121006/20241220_161728.jpg?ex=67673e86&is=6765ed06&hm=96e77f25a252a7dc4548b56c69974301610614b9364a27c623e0e641925dc7b4&
23:18 < bridge> <totar> I don't see how anyone will be writing code by hand in 8 years
23:19 < bridge> <Jupstar ✪> how big are the code tasks? 😄
23:19 < bridge> <totar> 2700 is top 0.01%
23:20 < bridge> <totar> Rank 150
23:20 < bridge> <inv41idu53rn4m3> Hey, more bugs to fix for us who do write by hand :D
23:20 < bridge> <inv41idu53rn4m3> :pepeW:
23:20 < bridge> <Jupstar ✪> but is that like codewars?
23:20 < bridge> <totar> It doesn't matter it crushes every benchmark
23:20 < bridge> <Jupstar ✪> yeah but software engineering isn't about _small_ algorithms
23:21 < bridge> <inv41idu53rn4m3> from what I've read about AI development it writes a ton of code that LGTM 👍 but actually has a bunch of bugs
23:21 < bridge> <totar> SWE bench the one to the left is about big problems
23:22 < bridge> <totar> It went from 1900 rating to 2700 rating in 3 months bro
23:22 < bridge> <totar> Its over
23:22 < bridge> <Jupstar ✪> before my AI can't code teeworlds on it's own, it's not AI xd
23:23 < bridge> <inv41idu53rn4m3> How does this rating work? Is it the number of problems solved correctly or does it take into account the performance of the solutions?
23:23 < bridge> <totar> I promise you're under estimating how much it can fit in the context window of new models
23:24 < bridge> <totar> This isn't the same thing as it was 6 months ago
23:24 < bridge> <inv41idu53rn4m3> Can you get a high Elo by solving just a few problems or do you need to solve a lot of them? I'm willing to bet that the 99.9% of users lower than that rating haven't actually solved most of the problems on the platform simply because they don't want to put in the hours
23:24 < bridge> <totar> I could dedicate my life and will not achieve 2700 rating
23:25 < bridge> <totar> IOI gold medal is 2400 rating
23:25 < bridge> <inv41idu53rn4m3> When a metric becomes a target it ceases to be a good metric
23:25 < bridge> <Jupstar ✪> i am open to be replaced by an AI.
23:25 < bridge> <Jupstar ✪> 
23:25 < bridge> <Jupstar ✪> But I am not really a to good to be true guy
23:25 < bridge> <Jupstar ✪> coding often means re-iterating over existing code
23:25 < bridge> <Jupstar ✪> and the current GPT models are not designed for such tasks
23:26 < bridge> <totar> You guys are missing the point, it's not longer about if it can do everything a software dev can it's about when. The science is over now it's just money and engineering.
23:26 < bridge> <totar> You guys are missing the point, it's no longer about if it can do everything a software dev can it's about when. The science is over now it's just money and engineering.
23:26 < bridge> <Jupstar ✪> I am currently not sure if you are serious xD
23:27 < bridge> <totar> I am totally serious 2700 cf is absurd
23:27 < bridge> <inv41idu53rn4m3> Also a robot being able to solve well established problems is not much more impressive than a robot rephrasing established literature. These things will still choke on novel problems because they lack proper ability to reason.
23:28 < bridge> <Jupstar ✪> chatgpt would probably easily beat me in A-Levels exams..
23:28 < bridge> <Jupstar ✪> 
23:28 < bridge> <Jupstar ✪> I'd still say it's stupid af xD
23:28 < bridge> <inv41idu53rn4m3> ok I'll trust you on that
23:28 < bridge> <inv41idu53rn4m3> but getting an absurd score on a bunch of microbenchmarks is not super relevant...
23:28 < bridge> <totar> I'm not going to pretend to be the chess players in 1960 who said a computer can never beat them at chess because it cannot beat the top 100 players
23:28 < bridge> <inv41idu53rn4m3> especially as it has probably read multiple articles about each of those problems
23:28 < bridge> <Jupstar ✪> I 100% believe AI is possible
23:28 < bridge> <inv41idu53rn4m3> these things have basically the entire internet in their training data
23:29 < bridge> <Jupstar ✪> But the way the current models work isn't really good for logical tasks
23:29 < bridge> <inv41idu53rn4m3> I absolutely believe proper intelligent AI is possible, but I do NOT believe we will reach anything like that by simply scaling up existing language models
23:29 < bridge> <totar> Current models you've used get 600 on code forces or worse probably
23:29 < bridge> <inv41idu53rn4m3> I haven't used any models hehe
23:30 < bridge> <inv41idu53rn4m3> How am I such a grumpy old man when it comes to computers? I've been like this since I've had a computer basically lol
23:30 < bridge> <From Europe> i guess im late since you already got the pip thing
23:30 < bridge> <totar> This is not a micro benchmark code forces was never intended to be an ai benchmark its competitive coding for humans
23:31 < bridge> <totar> It existed before chatgpt or llms
23:31 < bridge> <inv41idu53rn4m3> no I mean competitive programming tasks are all microbenchmarks
23:31 < bridge> <inv41idu53rn4m3> it's actually testing the speed of a single algorithm
23:32 < bridge> <From Europe> https://github.com/ddnet/ddnet/pull/7034#issuecomment-1700805235 but nonetheless
23:32 < bridge> <totar> At 2700 level it's not about speed alone
23:32 < bridge> <totar> You've never read a 2700 question i promise
23:33 < bridge> <inv41idu53rn4m3> Isn't that type of competitive programming *always* about the performance of the solution code?
23:33 < bridge> <inv41idu53rn4m3> or, well, assuming you can produce code that solves the problem in the first place
23:33 < bridge> <totar> past a point it becomes about solving the problem at all
23:34 < bridge> <totar> If youre not good at competitive programming then the questions are about optimizing an easy algorithm
23:34 < bridge> <inv41idu53rn4m3> I've competed at a national level but that's not really impressive considering the size of the nation lol
23:35 < bridge> <From Europe> but wouldn't it also be possible that they just have better training dataset now? could still fall apart as soon as new problem arises.
23:35 < bridge> <inv41idu53rn4m3> but at least I can say that I was one of like five people there who *weren't* from a fancy maths focused high school
23:35 < bridge> <From Europe> i've heard somewhere that AI benchmarks are basically just cheating dataset. forgot where i've heard that tho.
23:35 < bridge> <totar> Code forces rating can only be obtained in live competitions with new questions afaik
23:36 < bridge> <inv41idu53rn4m3> yeah these things have read basically the entire internet like three times over
23:36 < bridge> <inv41idu53rn4m3> (three hundred?)
23:36 < bridge> <inv41idu53rn4m3> (thousand?)
23:36 < bridge> <inv41idu53rn4m3> honestly no clue with how ridiculous the scale of computation is at this point
23:37 < bridge> <inv41idu53rn4m3> if there are solutions to these problems online these models have read them
23:37 < bridge> <Jupstar ✪> Hi @tsfreddie
23:38 < bridge> <From Europe> but like, how new tho. chinese olympiads trainees can also just do problems 24/7 for 6 years and get through a lot of problems not even understanding what they've learned and what algo actually uses in real applications.
23:39 < bridge> <totar> There's nothing they can say that will convince you then, they could just write any number they want on the graph if they wanted. I promise you the researchers making $5million a year are aware of dataset leakage
23:39 < bridge> <From Europe> prob now 2700 level but still
23:39 < bridge> <From Europe> prob not 2700 level but still
23:40 < bridge> <From Europe> Hi juppy :heartw:
23:40 < bridge> <inv41idu53rn4m3> I mean yeah, it's probably really good at some things, and definitely superhuman at many, but being superhuman at a few development related tasks doesn't really make you a software developer
23:41 < bridge> <totar> You won't get to experience these models first hand for a couple years because they are too expensive until improvements to hardware or algorithm is made so you can read the benchmarks and believe them or be surprised in 2 years when they aren't lying
23:41 < bridge> <From Europe> i knew actual people who just memorize like tons of snippets and just do problems with pattern recognitions :justatest:
23:42 < bridge> <inv41idu53rn4m3> This just sounds like solving rubik's cubes
23:42 < bridge> <From Europe> that's how a lot of us can skip chinese college entry exams
23:42 < bridge> <From Europe> i was in the same trianing camp since middle school
23:43 < bridge> <From Europe> but i failed tho, so mom sent me to the US since i've only done "coding" without going to actual schools for like 5 years
23:43 < bridge> <totar> The benchmark next to it is called "software engineer benchmark" chatgpt 4o gets 6% o3 gets 71% :cat_tired:
23:44 < bridge> <From Europe> where was oW
23:44 < bridge> <From Europe> where was o2
23:44 < bridge> <totar> Skipped because there's a company called o2
23:44 < bridge> <From Europe> :justatest:
23:45 < bridge> <inv41idu53rn4m3> I guess I should just become a carpenter
23:45 < bridge> <inv41idu53rn4m3> (mom also tells me that from time to time lol)
23:45 < bridge> <totar> The guy who invented the algorithm for o1 is saying this as well
23:45 < bridge> <totar> https://x.com/polynoamial/status/1870172996650053653?t=xlrABqsLiH6yLdsyX1hFMw&s=19
23:46 < bridge> <From Europe> o1 has 1891 is pretty cool too
23:46 < bridge> <inv41idu53rn4m3> Now the real question is if it's breaking the established scaling laws or not
23:46 < bridge> <From Europe> too bad im still using supermaven which is usually dumb af if you let it writing a thing from scratch
23:47 < bridge> <inv41idu53rn4m3> because that'll really tell you whether "this trajectory will continue" or not
23:47 < bridge> <totar> You can just make ASIC and it becomes 1000-2000x faster
23:47 < bridge> <From Europe> do they still do papers for o models
23:48 < bridge> <inv41idu53rn4m3> no but there are scaling laws for quality of output too, everyone kinda expects a plateau
23:48 < bridge> <totar> they don't tell you much
23:48 < bridge> <From Europe> too bad
23:48 < bridge> <inv41idu53rn4m3> (everyone except the VCs lol)
23:48 < bridge> <totar> 1900->2700 doesn't look like a plateau to me
23:49 < bridge> <From Europe> but again, i still believe if you seen enough you can get good at codeforces without understanding a lot of it
23:49 < bridge> <motkote> who understands, please help me how to compile the source code of the game via MSVS? I downloaded and changed the code that I need, BUT I DON'T UNDERSTAND HOW TO COMPILE?!?!?!?
23:49 < bridge> <inv41idu53rn4m3> That's why I'm asking if it's breaking the established scaling laws or not
23:49 < bridge> <inv41idu53rn4m3> because that's the deciding thing for me on how I'm viewing this 1900->2700 improvement
23:50 < bridge> <totar> Scaling laws only apply to pre training
23:50 < bridge> <totar> This is not that
23:50 < bridge> <totar> Scaling laws say nothing about RL training either
23:50 < bridge> <From Europe> is o3 the diffusion-adjacent thing or they don't tell you like at all
23:51 < bridge> <totar> Its a transformer
23:51 < bridge> <totar> As usual
23:51 < bridge> <inv41idu53rn4m3> transformers are so blursed
23:51 < bridge> <From Europe> i know but there was a non linear transformer thing iirc
23:51 < bridge> <From Europe> that doesn't do next token only
23:52 < bridge> <totar> They're not going to tell you anything more detailed other than "we trained it to reason"
23:52 < bridge> <From Europe> i guess so
23:53 < bridge> <From Europe> i can see generation quality get better if the AI doesn't need to cope with already generated nonsense due to bad luck on tempreture alone tho
23:53 < bridge> <totar> Can you get good at chess without understanding it if you just play thousands of games and memorize the patterns?
23:54 < bridge> <From Europe> oh o3 plays chess now?
23:54 < bridge> <From Europe> that's cool
23:54 < bridge> <totar> It did that since gpt4 turbo
23:55 < bridge> <From Europe> but like is it codeforces level of good
23:55 < bridge> <ryozuki> hi from europe
23:55 < bridge> <ryozuki> im from europe
23:55 < bridge> <From Europe> hi
23:55 < bridge> <totar> We already have chess robots better than humans
23:55 < bridge> <From Europe> im from europe too
23:55 < bridge> <inv41idu53rn4m3> Does it get its ass kicked by Stockfish?
23:55 < bridge> <totar> If it was better than humans at chess you would just say it's not impressive because it can't beat chess engines
23:56 < bridge> <inv41idu53rn4m3> I literally just said that lol
23:56 < bridge> <From Europe> i doubt stockfish is transformer
23:56 < bridge> <totar> Meanwhile it does your job because you job is not chess
23:56 < bridge> <Jupstar ✪> he works for openai btw
23:56 < bridge> <Jupstar ✪> 
23:56 < bridge> <Jupstar ✪> so if it's true what he says, and this is as impressive as it sounds, then gg
23:56 < bridge> <Jupstar ✪> 
23:56 < bridge> <Jupstar ✪> else it was just marketing and you fall for it
23:56 < bridge> <Jupstar ✪> 
23:56 < bridge> <Jupstar ✪> xd
23:56 < bridge> <totar> Meanwhile it does your job because your job is not chess
23:56 < bridge> <totar> I know he works for openai, he is the lead researcher who designed o1
23:57 < bridge> <totar> It was his idea
23:57 < bridge> <motkote> developers please help
23:57 < bridge> <motkote> hom compile game source?
23:57 < bridge> <Jupstar ✪> why
23:58 < bridge> <inv41idu53rn4m3> but yeah I guess I do need to consider carpentry or something if I can't make it in the business without dealing with this garbage
23:58 < bridge> <Jupstar ✪> read cpp tutorials and you'll understand
23:58 < bridge> <From Europe> and read this https://github.com/ddnet/ddnet/blob/master/README.md
23:59 < bridge> <From Europe> i mean i'm pretty sure coders are losing jobs regardless how good AI gets, cuz the business people are pretty much convinced already.
23:59 < bridge> <totar> I only posted his message because I trust him more than any of the other openai employees to be honest about his own algorithm