2010-12-03T00:02:41 *** Mathnerd314 has quit IRC (Ping timeout: 245 seconds) 2010-12-03T00:04:58 *** Appleman1234 has joined #aichallenge 2010-12-03T00:05:58 i'm surprised he didn't try to do something to the server while the contest ran 2010-12-03T00:06:21 try to get 4chan to ddos it 2010-12-03T00:06:25 o.o 2010-12-03T00:06:52 antimatroid: but then he would be the chiting one... he would become what he hates! 2010-12-03T00:07:06 :P 2010-12-03T00:07:32 hahaha hidden canadian reasons 2010-12-03T00:07:47 *** JnamedCndn has joined #aichallenge 2010-12-03T00:07:58 oh look, it's him again 2010-12-03T00:08:24 ohhi 2010-12-03T00:08:54 hey a1k0n, what happened to you during hte contest? 2010-12-03T00:08:58 you disappeared? 2010-12-03T00:09:11 yeah i kinda got discouraged after my ten different ideas all sucked and i had other work to do 2010-12-03T00:09:27 ah, yeah... I know the feeling :) 2010-12-03T00:09:34 minecraft, rock band 3, doing a bunch of neat machine learning stuff at work 2010-12-03T00:09:40 and what is krokokok's obsession with decision problems? 2010-12-03T00:09:40 i had plenty on my plate 2010-12-03T00:09:59 antimatroid: where's this? 2010-12-03T00:10:04 http://ai-contest.com/forum/viewtopic.php?f=19&t=1192&start=10 2010-12-03T00:10:28 a1k0n: yeah, this is a huge timesink... I essentially pulled out in the last month too 2010-12-03T00:10:55 * antimatroid wasted far too much time for a fairly meh finish 2010-12-03T00:11:23 i made this too: http://paste.lisp.org/display/117287 2010-12-03T00:11:31 antimatroid: you did pretty well 2010-12-03T00:11:44 sigh: after what i was expecting in tron i wanted that :P 2010-12-03T00:11:45 i probably would have worked harder if it were a shorter contest 2010-12-03T00:11:48 haha 2010-12-03T00:12:36 *** JnamedCndn has quit IRC (Ping timeout: 265 seconds) 2010-12-03T00:12:38 hmm... was krokokok the guy who was saying in irc that the contest not have games where players play against each other (a while back) 2010-12-03T00:12:46 yes 2010-12-03T00:12:47 pretty sure 2010-12-03T00:12:51 a1k0n: yes, I agree 2010-12-03T00:14:09 it's funny the sense of entitlement some people have, especially when the idea is such a huge departure from what the contest is about 2010-12-03T00:14:25 haha yep, but it's the only way you see :P 2010-12-03T00:14:30 :) 2010-12-03T00:15:02 i'd hate to see him solving a real world problem 2010-12-03T00:15:32 nah, it's too hard to do that properly, so let's just move on to something else 2010-12-03T00:15:47 a1k0n: hey, that program is cool, have you release the source? 2010-12-03T00:15:57 hehe 2010-12-03T00:16:27 has anyone seen the log of this channel that was taken for the article? 2010-12-03T00:16:41 epona: what do you mean? 2010-12-03T00:17:00 apparently some people were quoted 2010-12-03T00:17:17 are you asking where the channel logs are? 2010-12-03T00:17:26 *** smloh1 has quit IRC (Ping timeout: 265 seconds) 2010-12-03T00:17:53 no 2010-12-03T00:18:29 who knows. I don't remember the name of the people doing the article, I was jw if someone did 2010-12-03T00:18:44 eburnette was the one writing the article 2010-12-03T00:19:05 *** antimatroid1 has joined #aichallenge 2010-12-03T00:19:07 *** JnagJ has joined #aichallenge 2010-12-03T00:19:20 article is here: http://www.zdnet.com/blog/burnette/hungarian-lisp-developer-walks-away-with-google-ai-contest/2131 2010-12-03T00:20:19 people can speak lisp! 2010-12-03T00:20:25 *** antimatroid has quit IRC (Ping timeout: 264 seconds) 2010-12-03T00:20:34 i'm not sure i'd like to have a programming language spoken to me :P 2010-12-03T00:20:59 if(speak(c++) { lol();} 2010-12-03T00:21:23 you missed a bracket :P 2010-12-03T00:21:28 bad nesting! 2010-12-03T00:21:35 i win! (that's a game we play in maths lectures 2010-12-03T00:21:44 ) 2010-12-03T00:21:45 crap, i did it too ehe 2010-12-03T00:21:56 heh, I thought it was on purpose 2010-12-03T00:22:04 this article seems wrong 2010-12-03T00:22:34 *** davidjliu has quit IRC (Quit: Page closed) 2010-12-03T00:22:46 epona: I agree to some extent... I feel like it misrepresents some things 2010-12-03T00:22:50 "The PlanetWars contest received extra attention due to a post on reddittitled “The end is near! Self-improving AI based on Genetic Programming is beating 95% of hand-coded submissions in the Google AI Challenge”. " 2010-12-03T00:22:50 the contest received any attention cause of reddit :) 2010-12-03T00:22:57 but that's journalism -_- 2010-12-03T00:23:23 eburnette was a cool enough guy though, so meh 2010-12-03T00:23:39 no, this was on /b/ as well 2010-12-03T00:23:50 ah okay 2010-12-03T00:24:03 antimatroid1: yeah, sure... nothing against him 2010-12-03T00:24:04 i expected /b/ to be mostly 13 year old kids these days :\ 2010-12-03T00:24:05 I made sure to advertise like every day 2010-12-03T00:24:16 some of it 2010-12-03T00:24:34 antimatroid1: I 2010-12-03T00:24:48 sigh:? you're a channer? 2010-12-03T00:24:52 I've got a 40yo co-worker who does on /b/ 2010-12-03T00:24:54 no I'm not 2010-12-03T00:25:02 i've never been able to get down with that crowd 2010-12-03T00:25:08 me neither 2010-12-03T00:25:09 and never really wanted to :P 2010-12-03T00:25:13 its a great way to spread news, is all im sayin 2010-12-03T00:25:17 same with reddit 2010-12-03T00:25:42 yeah, they've done some cool stuff, but it seems to be more about trolling each other with disgusting photos 2010-12-03T00:26:28 the best along with the worst crowd 2010-12-03T00:26:45 sigh: not yet, but i will 2010-12-03T00:27:27 a1k0n: you're example all use standard words, how does it fare with bad spelling, name and such? 2010-12-03T00:27:44 give me an example 2010-12-03T00:27:55 i mean, yeah, it isn't great at names 2010-12-03T00:28:02 meh, take any of Accoun's posts :P 2010-12-03T00:28:05 i trained it on my ebook collections 2010-12-03T00:28:09 a1l0n have you tried mapping things like isnt to isn't? 2010-12-03T00:28:10 now that's a challenge. hmm 2010-12-03T00:28:19 nope, not yet 2010-12-03T00:28:49 one could thing would be to branch a tree of possible word cut offs? then try to evalutate terminal setences as to which is most likely valid? 2010-12-03T00:28:56 cool* thing 2010-12-03T00:29:31 *** JnagJ has quit IRC (Ping timeout: 265 seconds) 2010-12-03T00:29:32 it uses a 6-character-gram model, and dynamic programming ot find the maximum likelihood parsing 2010-12-03T00:29:35 runs in linear time 2010-12-03T00:29:36 i have no idea how one would evaluate that, validity of the grammer stuff etc. perhaps? 2010-12-03T00:29:50 so if you can come up with a better language model, it'd use it 2010-12-03T00:29:53 ah ok, makes sense 2010-12-03T00:30:11 so is the branching technique in coding the superior way to make educated decisions? 2010-12-03T00:30:22 what's kinda cool is the language model is hashed into (insert size here) space, with a pretty smooth degradation of performance as you shrink down the weights 2010-12-03T00:30:24 that tree thing would suck with large blocks of text,but could possibly be run on individual sentences? 2010-12-03T00:30:34 the branching technique? 2010-12-03T00:30:36 antimatroid1: look into markov-chains 2010-12-03T00:30:42 that's what a1k0n is using 2010-12-03T00:30:45 i know what markov chains are :P 2010-12-03T00:30:56 why don't people just say that :P 2010-12-03T00:30:59 yeah, i've implemented that sort of thing before. you can do dynamic programming on (tree context, rest of string) 2010-12-03T00:31:15 Viterbi algorithm, I think? 2010-12-03T00:31:18 but then you absolutely have to have every word in your dictionary 2010-12-03T00:31:18 is that right? 2010-12-03T00:31:28 yeah, it might be equivalent to viterbi, i hadn't thought of that 2010-12-03T00:31:44 burny:the technique bocsimacko used 2010-12-03T00:31:48 what does it do now? pick words then make words out of anything left? 2010-12-03T00:32:29 given the (up to) five letters in a word, what's the likelihood of the letter under consideration? what's the likelihood of a word break here, and then starting a new word from here? 2010-12-03T00:32:36 five is an adjustable parameter of course 2010-12-03T00:32:50 the training is done on my collection of plaintext ebooks for my kindle.. 2010-12-03T00:33:02 train it on the channel logs :P 2010-12-03T00:33:14 heh. could do that. 2010-12-03T00:34:12 anyway it's a neat little algorithm. nothing new, just kinda cool that it runs in linear time. 2010-12-03T00:34:21 :) 2010-12-03T00:34:54 a1k0n, may I see it? 2010-12-03T00:35:13 -39.9 again we have hidden can a dian resons 2010-12-03T00:35:30 haha 2010-12-03T00:36:20 what is that number? 2010-12-03T00:36:34 proportional to log of probablity? 2010-12-03T00:36:42 yeah, precisely. per-character. 2010-12-03T00:36:47 -80.7 c jnvxvkjz nvlk sadnjvkldnxv 2010-12-03T00:36:51 so you can tell whether it's english or not 2010-12-03T00:36:55 (roughly) 2010-12-03T00:36:57 ah, cool 2010-12-03T00:37:01 http://paste.lisp.org/display/117288 2010-12-03T00:37:02 that's the code 2010-12-03T00:37:08 for the segmentation anyway 2010-12-03T00:37:19 anyway my wife wants to play monopoly on the xbox now 2010-12-03T00:37:25 ok 2010-12-03T00:37:28 cya :) 2010-12-03T00:37:59 sigh: did you see the trapezoidal post on reddit 2010-12-03T00:38:01 (my ctx_likelihood function only takes a-z right now) 2010-12-03T00:38:06 antimatroid1: no 2010-12-03T00:38:25 http://www.reddit.com/r/math/comments/eero3/1_write_scientific_paper_on_using_the_trapezoidal/ 2010-12-03T00:38:33 ah, the chem paper? 2010-12-03T00:38:39 yeah, I did see it :P 2010-12-03T00:38:48 the article it sites is almost even more golden 2010-12-03T00:38:57 i was laughing like a little girl for a good 5 minutes 2010-12-03T00:39:04 cites* 2010-12-03T00:39:57 quiet the long list :) 2010-12-03T00:41:47 *** werwer has joined #aichallenge 2010-12-03T00:42:22 with a revolutionary technique like that, what do you expect :P 2010-12-03T00:42:58 "Can't wait until they discover Simpsons Rule." haha 2010-12-03T01:09:47 reddit is so broken lately 2010-12-03T01:09:54 when i finally have a need for it again 2010-12-03T01:11:23 at least it means I can get work done :P 2010-12-03T01:11:46 i have no work to get done :P 2010-12-03T01:15:04 ai :P 2010-12-03T01:15:14 for that chess game, or whatever 2010-12-03T01:15:26 *** werwer has quit IRC (Ping timeout: 265 seconds) 2010-12-03T01:15:27 nah, i can't bear to write another minimax tree right now 2010-12-03T01:15:33 i'll do that slowly over a year 2010-12-03T01:15:45 make a maxmax tree then :P 2010-12-03T01:16:11 or try out UCT :P 2010-12-03T01:16:20 it would make a lot more sense to call them maximin trees 2010-12-03T01:16:34 UCT is just used for searching said tree yeah? 2010-12-03T01:16:55 i will start working on my game theory library again at some point, i at least get to use proper maths for that 2010-12-03T01:17:10 but it gets really hard fast too :S 2010-12-03T01:17:42 UCT is used for the same thing as minimax 2010-12-03T01:18:25 it's too much of the same thing over and over 2010-12-03T01:18:37 :P 2010-12-03T01:19:14 hey, did you ever end up using simplex for anything in this contest, like you said you would try 2010-12-03T01:19:41 no, wouldn't have worked fast enough for me 2010-12-03T01:19:52 i wanted to use it for move selection 2010-12-03T01:20:05 or finding mixed strategy nash equilibrium in a zero sum simultaneous game 2010-12-03T01:20:12 fair enough 2010-12-03T01:20:34 give the bots 10 seconds and they could be a lot better 2010-12-03T01:21:38 don't like that, the games become too long so harder to test and can't play as many games 2010-12-03T01:21:48 yeah i don't want to do it 2010-12-03T01:22:55 if i had more time, i would probably use decision tree's at each node inside the minimax tree to pick moves for each player, and actually increment states 2010-12-03T01:29:27 *** sigh has left #aichallenge ("Leaving") 2010-12-03T01:34:28 *** Utkarsh has quit IRC (Ping timeout: 250 seconds) 2010-12-03T01:35:39 *** tapwater has quit IRC (Quit: tapwater) 2010-12-03T01:39:49 *** Utkarsh has joined #aichallenge 2010-12-03T01:49:24 interesting, can you guys access wikileaks.org? 2010-12-03T01:49:35 people on my facebook started complaining, and it seems I can't either 2010-12-03T01:57:10 *** iFire has quit IRC (Ping timeout: 255 seconds) 2010-12-03T01:59:10 *** Appleman1234 has quit IRC (Remote host closed the connection) 2010-12-03T02:04:26 *** iNtERrUpT has quit IRC (Remote host closed the connection) 2010-12-03T02:04:56 *** iFire has joined #aichallenge 2010-12-03T02:06:47 *** Appleman1234 has joined #aichallenge 2010-12-03T02:20:06 *** amstan has joined #aichallenge 2010-12-03T02:20:06 *** ChanServ sets mode: +o amstan 2010-12-03T02:26:01 *** iFire has quit IRC (Read error: Connection reset by peer) 2010-12-03T02:26:22 *** iFire has joined #aichallenge 2010-12-03T02:33:37 *** sigh has joined #aichallenge 2010-12-03T02:43:09 *** antimatroid has joined #aichallenge 2010-12-03T02:44:07 *** antimatroid1 has quit IRC (Ping timeout: 240 seconds) 2010-12-03T02:55:39 *** Azrathud has quit IRC (Ping timeout: 255 seconds) 2010-12-03T03:25:07 *** mega1 has quit IRC (Ping timeout: 240 seconds) 2010-12-03T03:29:48 *** narnach1 has joined #aichallenge 2010-12-03T03:29:48 *** narnach1 is now known as narnach_mb 2010-12-03T03:54:49 *** aerique has joined #aichallenge 2010-12-03T04:00:15 *** meneky_ has quit IRC (Ping timeout: 265 seconds) 2010-12-03T04:20:16 *** Appleman1234 has quit IRC (Ping timeout: 240 seconds) 2010-12-03T04:32:47 *** nullkuhl has quit IRC (Ping timeout: 240 seconds) 2010-12-03T04:37:00 *** Naktibalda has joined #aichallenge 2010-12-03T04:39:04 *** sigh has quit IRC (Remote host closed the connection) 2010-12-03T04:49:06 *** nullkuhl has joined #aichallenge 2010-12-03T04:56:08 *** jaspervdj has joined #aichallenge 2010-12-03T05:01:24 *** sigh has joined #aichallenge 2010-12-03T05:04:59 *** Appleman1234 has joined #aichallenge 2010-12-03T05:14:16 *** mceier has joined #aichallenge 2010-12-03T05:14:18 *** medrimonia1 has quit IRC (Quit: Leaving.) 2010-12-03T05:23:42 *** iamluck|2 has quit IRC (Ping timeout: 255 seconds) 2010-12-03T05:24:24 *** bhasker has joined #aichallenge 2010-12-03T06:11:27 *** nullkuhl has quit IRC (Ping timeout: 240 seconds) 2010-12-03T06:12:20 *** Frontier has joined #aichallenge 2010-12-03T06:15:53 *** Sylph has quit IRC (Read error: Connection reset by peer) 2010-12-03T06:16:02 *** Sylph has joined #aichallenge 2010-12-03T06:30:52 *** nullkuhl has joined #aichallenge 2010-12-03T06:34:31 *** nullkuhl has quit IRC (Read error: Connection reset by peer) 2010-12-03T06:36:38 *** nullkuhl has joined #aichallenge 2010-12-03T06:42:49 *** amstan has quit IRC (Ping timeout: 255 seconds) 2010-12-03T06:46:51 *** bhasker has quit IRC (Quit: bhasker) 2010-12-03T06:54:22 *** dmj111 has quit IRC (Quit: ERC Version 5.3 (IRC client for Emacs)) 2010-12-03T07:00:24 *** amstan has joined #aichallenge 2010-12-03T07:00:24 *** ChanServ sets mode: +o amstan 2010-12-03T07:57:25 *** mega1 has joined #aichallenge 2010-12-03T07:57:43 *** phreeza has joined #aichallenge 2010-12-03T08:08:19 *** Mistmanov has joined #aichallenge 2010-12-03T08:08:57 *** Erika_ has joined #aichallenge 2010-12-03T08:09:21 *** epona has quit IRC (Read error: Connection reset by peer) 2010-12-03T08:18:07 *** meneky_ has joined #aichallenge 2010-12-03T08:21:57 *** p4p4 has joined #aichallenge 2010-12-03T08:33:21 *** meneky_ has quit IRC (Quit: Page closed) 2010-12-03T08:44:07 *** Mistmanov has quit IRC (Quit: Page closed) 2010-12-03T08:51:08 *** yasith has quit IRC (Ping timeout: 240 seconds) 2010-12-03T09:04:13 *** yasith has joined #aichallenge 2010-12-03T09:08:53 *** Sylph2 has joined #aichallenge 2010-12-03T09:09:01 *** Sylph has quit IRC (Read error: Connection reset by peer) 2010-12-03T09:12:24 *** Sylph has joined #aichallenge 2010-12-03T09:15:40 *** Sylph2 has quit IRC (Ping timeout: 260 seconds) 2010-12-03T09:19:44 *** Sylph2 has joined #aichallenge 2010-12-03T09:21:27 *** Sylph has quit IRC (Ping timeout: 240 seconds) 2010-12-03T09:24:51 *** ebrahim has joined #aichallenge 2010-12-03T09:24:51 *** ebrahim has joined #aichallenge 2010-12-03T09:27:20 *** sigh has quit IRC (Remote host closed the connection) 2010-12-03T09:27:45 *** Erika_ has quit IRC (Ping timeout: 272 seconds) 2010-12-03T09:30:39 *** narnach has quit IRC (Ping timeout: 265 seconds) 2010-12-03T09:32:53 *** mega1 has quit IRC (Remote host closed the connection) 2010-12-03T09:33:30 *** mega1 has joined #aichallenge 2010-12-03T09:35:43 *** Erika_ has joined #aichallenge 2010-12-03T09:47:21 *** Erika_ has quit IRC (Ping timeout: 276 seconds) 2010-12-03T09:48:59 *** Rubicon-|-Cross has quit IRC (Quit: Leaving) 2010-12-03T09:50:35 *** meneky_ has joined #aichallenge 2010-12-03T09:52:14 *** Erika_ has joined #aichallenge 2010-12-03T09:54:39 *** Appleman1234 has quit IRC (Ping timeout: 240 seconds) 2010-12-03T09:57:34 *** Appleman1234 has joined #aichallenge 2010-12-03T10:00:07 *** Erika_ has quit IRC (Ping timeout: 240 seconds) 2010-12-03T10:08:51 *** RainCT has joined #aichallenge 2010-12-03T10:11:25 *** nullkuhl has quit IRC (Read error: Connection reset by peer) 2010-12-03T10:12:10 *** bhasker has joined #aichallenge 2010-12-03T10:13:09 *** iamluck has joined #aichallenge 2010-12-03T10:14:07 *** nullkuhl has joined #aichallenge 2010-12-03T10:17:14 *** contestbot_ has joined #aichallenge 2010-12-03T10:17:14 *** contestbot has quit IRC (Disconnected by services) 2010-12-03T10:17:55 *** amstan_ has joined #aichallenge 2010-12-03T10:17:55 *** ChanServ sets mode: +o amstan_ 2010-12-03T10:18:02 @contestbot: You must change your name :) 2010-12-03T10:18:03 Frontier: Run as fast as you can and don't look back. 2010-12-03T10:18:47 *** amstan has quit IRC (Ping timeout: 240 seconds) 2010-12-03T10:18:47 *** contestbot_ is now known as contestbot 2010-12-03T10:18:47 *** ChanServ sets mode: +o contestbot 2010-12-03T10:21:10 maybe already posted here, Mistmanov posted in the forum what i think is the best game idea so far http://ai-contest.com/forum/viewtopic.php?f=21&t=1205&start=0 2010-12-03T10:22:49 *** Sylph has joined #aichallenge 2010-12-03T10:22:52 * antimatroid takes a read 2010-12-03T10:22:54 Oh nice, that is closely related to my ideas based on the old RoboCom game: each robot/and runs it's own copy of the code. Does the code run as Python/C/Haskel/... or is it a virtual machine? 2010-12-03T10:23:28 I mean - it doesn't run at all right now, but what was Mistmanov's plan? 2010-12-03T10:23:45 *** boegel has joined #aichallenge 2010-12-03T10:24:21 any word on the next challenge? :) 2010-12-03T10:24:45 i think there's a fairly clear trend going on in waht people want 2010-12-03T10:24:58 multiple bots from each player trying to achieve a common goal 2010-12-03T10:25:17 woa, cool 2010-12-03T10:25:19 boelel: http://ai-contest.com/forum/viewtopic.php?f=21&t=1205&start=0 2010-12-03T10:25:27 nothing decided, just discussions 2010-12-03T10:25:32 that's boeGel, but thx :) 2010-12-03T10:25:43 yes, sorry that was just a typo :) 2010-12-03T10:26:31 you wont hear anything concrete till at least early next year 2010-12-03T10:28:06 what's the time frame for the next contest? 2010-12-03T10:29:53 *** Migi32 has joined #aichallenge 2010-12-03T10:29:56 *** Migi32 has left #aichallenge 2010-12-03T10:30:30 it's not really set 2010-12-03T10:30:35 tron was in february 2010-12-03T10:31:06 *** narnach has joined #aichallenge 2010-12-03T10:32:52 i would make the radius of vision for each bot discrete 2010-12-03T10:33:07 like you can see a bot that is on a square with a min path to is <= x 2010-12-03T10:34:37 hmmm what about everybody edits, but you both start at the bottom and it's a race to the top? 2010-12-03T10:37:05 http://everybodyedits.com/ 2010-12-03T10:38:07 how about a second, lager radius that only gives you the information that there IS a drone, but now which team it is on? :) 2010-12-03T10:38:07 Damn I would give all my drones clear orders so they all still know where they will linger after a few turns. Nothing worse than finding an enemy drone and not knowing where the support is. 2010-12-03T10:40:32 once you know where two are in relation to each other, you can store a said information and track where they are in relation to each other 2010-12-03T10:40:35 that can't be prevented 2010-12-03T10:40:51 i'm not really sure how to do it with a grid either 2010-12-03T10:41:06 if you provide details of your bots locations, you know where they are in relation to each other 2010-12-03T10:44:46 *** nullkuhl has quit IRC (Ping timeout: 245 seconds) 2010-12-03T11:16:00 *** bhasker has quit IRC (Quit: bhasker) 2010-12-03T11:30:39 *** migi_ has joined #aichallenge 2010-12-03T11:31:00 amstan_: why would programs be short? That's a rather bold statement :P 2010-12-03T11:35:02 *** aerique has quit IRC (Quit: ...) 2010-12-03T11:36:22 *** nullkuhl has joined #aichallenge 2010-12-03T11:37:52 Error323: i think he meant the same with this 2010-12-03T11:37:57 in that they couldn't do very much 2010-12-03T11:38:45 *** sow has joined #aichallenge 2010-12-03T11:39:54 *** holmar has quit IRC (Ping timeout: 255 seconds) 2010-12-03T11:40:25 couldn't do very much? 2010-12-03T11:40:45 with 1 second and the number of possible moves, not really :P 2010-12-03T11:44:04 *** Migi32 has joined #aichallenge 2010-12-03T11:44:09 *** Mistmanov has joined #aichallenge 2010-12-03T11:49:48 Mistmanov: cool idea :) 2010-12-03T11:49:58 thanks =) 2010-12-03T11:50:08 you do realize its RTS? ;) 2010-12-03T11:50:14 ah, not really :p 2010-12-03T11:50:31 well it is :P 2010-12-03T11:50:32 rts is more one supreme commander giving all the commands 2010-12-03T11:50:37 in my view anyway 2010-12-03T11:50:44 this is more simulating emergent behavior as found in nature 2010-12-03T11:50:45 *** holmar has joined #aichallenge 2010-12-03T11:50:47 completely different ;) 2010-12-03T11:51:19 well in this case the ai programmer is the supreme commander :P 2010-12-03T11:51:26 biggest problem would be getting the parameters right so they allow interesting beneficial tactics 2010-12-03T11:51:45 if you get the parameters wrong, there's a big chance that full on defense of full on offense is the ideal solution 2010-12-03T11:52:10 *or full on offense 2010-12-03T11:52:37 yes it needs a good balance 2010-12-03T11:52:47 Error323: but comm restriction makes the commander not to be absolutely on command :P , i agree mistmanov, more like an emergent behaviour game, and i really like the idea 2010-12-03T11:52:54 what you want is that you need to band together for defense, but that raiding parties still have a chance of slipping in to destroy eggs if the perimeter is defended poorly etc 2010-12-03T11:53:59 I'd like to say that while I rather like the idea of multiple independent agents for the player, trying to run a player's program every time for each agent with potentially many, many agents is going to end up a nightmare 2010-12-03T11:54:19 yeah, that's the big disadvantage 2010-12-03T11:54:43 something like 100 drones per player would be cool, but then you can't do 1 sec per drone per turn 2010-12-03T11:55:14 (and I don't know enough about computers to know if running 200 programs at the same time is a problem anyway) 2010-12-03T11:55:45 well on 100 per player its a lot more 2010-12-03T11:56:21 but with sth like 5 or 10 as max, interesting behavior and a real challenge is already created 2010-12-03T11:56:34 I should write a post just describing what would be good features for a game from a technical point of view from what I'm seeing with the current contest 2010-12-03T11:56:35 *** holmar has quit IRC (Ping timeout: 272 seconds) 2010-12-03T11:57:00 e.g. things like games that generate less data, good. games that are short, good. :) 2010-12-03T11:57:35 *** holmar has joined #aichallenge 2010-12-03T11:57:51 well it depends on the quality of a single game win/loss 2010-12-03T11:58:13 if thats really high then much less games are required for scoring 2010-12-03T11:58:21 would it help if the players just wrote a function that handles all the stuff, and then the server runs a single program that just executes that function once for every drone? 2010-12-03T11:59:11 how do you prevent the player from having global state? 2010-12-03T11:59:15 i you run 2 or 3 games for each player simultaneusly you can spawn x processes per game and player, and call each process with the a random drone of any of the 3 games, with that you can almost ensure that no global info can be stored (restricted comm), the only problem is that you cannot keep info from turn to turn. 2010-12-03T12:00:17 shouldn't a drone be able to have persistent memory throughout the game? 2010-12-03T12:00:24 yeah I think they should 2010-12-03T12:00:39 that way if you want to calculate something really complicated 2010-12-03T12:00:49 you could just decide to just calculate for a few turns 2010-12-03T12:01:31 I think between persistent memory and not wanting 'side-band' communication among the drones you would be forced to have separate programs for each drone :( 2010-12-03T12:02:00 hmm :( 2010-12-03T12:02:08 yep,it looks like too heavy load 2010-12-03T12:02:25 convince google to sponsor you with giant datacenters 2010-12-03T12:02:28 *** boegel has quit IRC (Quit: Leaving) 2010-12-03T12:02:42 instead of just name -.- 2010-12-03T12:03:36 Error323: my understanding now is that it turns out google does supply money as well although I don't know what sort of budget 2010-12-03T12:03:51 oh cool 2010-12-03T12:04:22 yeah, I was happy to learn that 2010-12-03T12:04:37 On the other hand.. forcing the drones to do all their calculations in short turns (say less than 100 ms) would also be a nice challenge 2010-12-03T12:04:51 but I don't know how much that puts some languages at a disadvantage 2010-12-03T12:05:03 100ms is very very little :/ 2010-12-03T12:05:08 a second is already very little 2010-12-03T12:05:29 100 ms is a huge interval in c++ :P 2010-12-03T12:05:32 true, but ants sure seem to do well with little processing power 2010-12-03T12:06:14 what about reducir further the number of drones (say 20) and reducing the time for each (say 100 ms per turn) at the same time? 2010-12-03T12:06:32 I don't know what kind of processing power they have tbh, but a few billion years of evolution also helped :P 2010-12-03T12:06:32 no new egg will convert to a drone if the maximun is reached 2010-12-03T12:06:49 janzert: can you give each player 1 core? 2010-12-03T12:06:56 and then run as many instances as necessary? 2010-12-03T12:07:02 that would be completely fair if feasible 2010-12-03T12:07:24 probably not 2010-12-03T12:07:25 the OS handles the scheduling 2010-12-03T12:07:38 ahhh, reading mega1's post-mortem kinda hurts. I did the "full attack based evaluation function" almost exactly the same as him, but when executing it it didn't perform well. It didn't occur to me that I could just let it ignore neutrals to make it work... 2010-12-03T12:08:29 Mistmanov: in your image, you went with a discrete map but continuous radius? :P 2010-12-03T12:08:38 wouldn't it be better to have the radius be discrete too 2010-12-03T12:08:44 yeah 2010-12-03T12:08:52 but contious radius was 1 click with a big brush 2010-12-03T12:08:53 :p 2010-12-03T12:08:59 :P 2010-12-03T12:09:03 if it wasn't for trying to include as many programming languages as possible I'd say have the drones run an artificial bytecode specific to the contest 2010-12-03T12:09:17 that the engine would directly interpret 2010-12-03T12:09:43 and the drone time slice can be so many bytecodes 2010-12-03T12:09:44 janzert: have you read Frontier's post on the forum? 2010-12-03T12:09:55 what about allowing all bots to communicate with each other (so they all run in the same program) but they onlsee see x distance from any friend bot 2010-12-03T12:10:00 janzert: that's a good idea 2010-12-03T12:10:03 all friend bots* 2010-12-03T12:10:10 ;] 2010-12-03T12:10:18 meneky_: not sure, which one? 2010-12-03T12:10:36 antimatroid: that's also possible, but then you kinda lose the point of the entire game :p 2010-12-03T12:10:43 antimatroid: that's basically RTS with fog of war right? 2010-12-03T12:10:44 that's more a classic rts 2010-12-03T12:11:10 janzert: this one http://ai-contest.com/forum/viewtopic.php?f=21&t=1191, i have just take a superficial llok at it 2010-12-03T12:11:20 how can you have turn based rts? 2010-12-03T12:11:26 i don't think i understand the meaning of rts 2010-12-03T12:11:35 ahh, no I hadn't 2010-12-03T12:12:19 for humans, the difference between rts and tbs is that in a tbs you have all the time in the world to think about your moves, while in rts you need to act quickly 2010-12-03T12:12:33 for programs, that difference falls away a bit imho 2010-12-03T12:12:43 isn't real time strategy implying time is continuous? 2010-12-03T12:12:45 not discrete 2010-12-03T12:13:04 that's what i had always gathered 2010-12-03T12:13:09 yeah but for a computer the time is always going to be discrete 2010-12-03T12:13:16 well no 2010-12-03T12:13:31 you can have no "turns" and just let bots execute moves as they want 2010-12-03T12:13:39 the max time for a game can be used 2010-12-03T12:13:57 that's an interesting thought 2010-12-03T12:14:05 well, the game engine would still only read input at specific timesteps right? 2010-12-03T12:14:07 antimatroid: I think of the difference between RTS and TBS is rts time advances whether the player does anything or not 2010-12-03T12:14:23 the RTS time is actually often discrete at the simulation level 2010-12-03T12:14:30 not necessarily; all unit actions can be defined in terms of continuous time 2010-12-03T12:14:34 each timestep that a bot gives no commands is then just a turn in which the bot gave no commands 2010-12-03T12:14:43 *** sow has quit IRC (Quit: Page closed) 2010-12-03T12:14:46 but generally it's implemented with timesteps, yes 2010-12-03T12:15:56 Mistmanov: sure i guess, but a lot of people don't really treat it like that 2010-12-03T12:16:09 i think a contest where there was no concept of turns at all would be interesting 2010-12-03T12:16:26 but how do you let the bots keep track of the gamestate then? If you let them track it themselves there can be rounding errors in the floating point operations, and updating it after every command can become quite cpu-intensive 2010-12-03T12:16:34 i think pursuier/evader games would work for that, although it really needs to have both as both 2010-12-03T12:16:58 ^^ hadn't got tht far :P 2010-12-03T12:18:18 but yea a "real time" game would be really cool. You don't have artificial computing time limits then, either. 2010-12-03T12:19:49 I think computing time limits would become way more important in a real time game 2010-12-03T12:20:10 if your bot takes 100 ms to adapt to a new situation, and the enemy bot is doing something radically different every 10 ms 2010-12-03T12:21:13 true. There'll be a real trade-off between how much you want your bot to calculate before doing his first move. 2010-12-03T12:21:17 say.. if in planetwars you could redirect your ships midflight as ofen as you want 2010-12-03T12:21:27 faster bots have a real advantage there 2010-12-03T12:24:04 hmmm. You could also just make the other bot crash by sending too many orders, though. 2010-12-03T12:24:24 Yup. 2010-12-03T12:25:55 so you should at least impose a limit on the number of orders per turn 2010-12-03T12:26:00 eh per second 2010-12-03T12:27:13 Migi32: I knew it would break someone's heart :-) 2010-12-03T12:28:06 take comfort in the fact that those MIN-TURN constraint are crucial as well. 2010-12-03T12:28:55 mega1, no I designed it with the min-turn stuff in mind 2010-12-03T12:29:27 well, in that case I can offer no consolation. 2010-12-03T12:30:02 aww :( 2010-12-03T12:30:25 but no worries, seeing how close I got to the best solution I won't hold back next time :P 2010-12-03T12:31:26 nobody was close :P 2010-12-03T12:31:35 getting the fine details perfect was very important too 2010-12-03T12:31:49 Migi32: I still feel that the move generator was underemphasized. 2010-12-03T12:31:58 i think the actual continous time idea is quite bad now 2010-12-03T12:32:13 mega1: i still think it's all in the move selection :P 2010-12-03T12:32:32 yes, that's what I mean 2010-12-03T12:35:16 antimatroid, well at one point I was fairly close, not in ELO but in general idea. But yea, when I realised that this kind of move evalution will prefer doing nothing so it still has many options left instead of being commited to one, I thought this was a fundamental flaw that couldn't be fixed. 2010-12-03T12:36:16 that's where min-turn should help 2010-12-03T12:36:25 most of the time 2010-12-03T12:36:44 also, I've read the code and (while I don't understand Lisp too much) the move generation was fairly similar as well 2010-12-03T12:37:43 there were about a hundred little variants, tweak of the move generator 2010-12-03T12:37:59 it was a very sensitive instrument 2010-12-03T12:38:01 i think a lot of people did similar things though :P i know i did in places 2010-12-03T12:38:11 it was the entire combination, and without a doubt he pantsed the lot of us 2010-12-03T12:38:28 well, accoun will doubt 2010-12-03T12:38:30 nobody else does 2010-12-03T12:38:33 lol 2010-12-03T12:38:39 :) 2010-12-03T12:38:55 One important lesson I learnt is how important it is to fix things in the right order. 2010-12-03T12:39:31 i'm not sure what i leant with mine other than do better :P 2010-12-03T12:39:42 but yea, I'm not saying I would've gotten first place. But after reading this I think top 10 was definitely possible 2010-12-03T12:39:53 heh i also did the full attack based eval function 2010-12-03T12:40:14 Migi32: I'm not trying to be an ass, but so did 50 other people :P 2010-12-03T12:40:27 i'm just an ass naturally 2010-12-03T12:40:36 a1k0n: did you use it for the neutrals too? 2010-12-03T12:40:41 yes. 2010-12-03T12:40:55 well, i mainly used to determine how many ships i can afford to send out 2010-12-03T12:41:01 but that was too conservative 2010-12-03T12:41:20 ah, yes. full-attack based move generation sucked 2010-12-03T12:41:42 that is what i originally did, but never worked out the issues with it 2010-12-03T12:41:42 I used the normal future for that. 2010-12-03T12:41:42 i didn't have the initial insight you did that if you can take one planet you can take them all 2010-12-03T12:41:54 *** narnach_mb has quit IRC (Quit: Leaving.) 2010-12-03T12:42:10 (or if you cannot take one then you can take none) 2010-12-03T12:42:18 right 2010-12-03T12:42:41 I haven't proved it though 2010-12-03T12:42:49 may be wrong. 2010-12-03T12:43:37 my move generator was basically to compute the future out to the horizon, and then work backward and determine how many ships are needed to take each planet from each side 2010-12-03T12:43:59 mega1: you assumed that if one planet can't be taken then another can't? 2010-12-03T12:44:02 then send out that many or more, as long as you don't weaken yourself by my other discounted-full-attack thing 2010-12-03T12:44:03 a1k0n: from each side? 2010-12-03T12:44:13 well, as two separate counts 2010-12-03T12:44:19 that is wrong in general, but probably not too often 2010-12-03T12:44:31 antimatroid: no, no. Read the full attack future section in the post-mortem. 2010-12-03T12:44:56 a1k0n: what did you need counts of the opponent for? 2010-12-03T12:45:02 so like, if player 2 had ships on the way to one of player 1's planets "i", planet[i].needed[1] would be the number of ships to fend off the attack 2010-12-03T12:45:08 mega1, talking about this bit? "if for all planets of player 1 player 2 cannot take that planet if both players sent all ships to it then player 2 cannot take any planet of player 1 even if he's allowed to attack multiple planets in any pattern" 2010-12-03T12:45:15 mega1: for minimax move generation 2010-12-03T12:45:18 Migi32: yes 2010-12-03T12:45:23 it's wrong 2010-12-03T12:45:31 I can give you a map with 4 planets to show you 2010-12-03T12:45:32 I'd not be surprised 2010-12-03T12:45:32 holdo n 2010-12-03T12:45:34 on* 2010-12-03T12:46:07 a1k0n: this is for simultaneous moves, right? 2010-12-03T12:46:11 (actually i had planets[i].frames[j].needed[1] rather) 2010-12-03T12:46:23 yes...? 2010-12-03T12:46:42 i didn't actually need a lot of the stuff i calculated, but i had it on hand with no clear idea of how best to exploit it. 2010-12-03T12:47:07 one idea that was too costly in terms of cpu time was to generate moves after observing good moves of the opponent 2010-12-03T12:47:11 oh, how far into the future did you make moves per planet? i just did it to maxDistance 2010-12-03T12:47:27 i did that initially 2010-12-03T12:47:35 and? 2010-12-03T12:47:38 but like, then it would make ridiculous assumptions about what the opponent would actually do 2010-12-03T12:48:03 oh, observing 2010-12-03T12:48:17 nevermind. you mean actually modelling the opponent? 2010-12-03T12:48:25 no, I think you answered what I had in mind. 2010-12-03T12:48:34 oh. yeah. i generated moves greedily for the opponent, then for myself 2010-12-03T12:48:38 then... i lost, and commented that out. 2010-12-03T12:49:06 it would try to snipe attacks that never happened, etc. 2010-12-03T12:49:13 might be useful when you have an advantage. dunno. 2010-12-03T12:49:20 ok, I actually wanted to generate moves for the top N opponent moves 2010-12-03T12:49:32 yeah, that could work but how do you decide which to use? 2010-12-03T12:49:40 which top N/ 2010-12-03T12:49:41 ? 2010-12-03T12:49:42 i guess use the moves that come out consistently? 2010-12-03T12:49:52 a1k0n: i solved the stupid snipes by not having either bot consider sending a snipable expansionary move 2010-12-03T12:50:01 the evaluation would stay the same 2010-12-03T12:50:03 which moves you make, given the best moves for you given the best top N opponent moves 2010-12-03T12:51:02 the best ones according to my normal evaluation function? 2010-12-03T12:51:07 antimatroid: yeah that would probably help. 2010-12-03T12:51:59 mega1: i guess what i'm saying is, if you assume in turn the enemy does N different things, and for each of those, pick a set of moves... how do you finally pick among the moves? your evaulation scores are each dependent on the opponent action 2010-12-03T12:52:16 also, sending ships at the very last possible turn meant that i could respond to their "potential moves" and not stupidly send stuff of for moves they made unless it was needed to cancel it 2010-12-03T12:52:18 so you'd get a score boost if you assume the opponent does something stupid, no? 2010-12-03T12:52:50 yeah i should have worked on move scheduling more. i kept implementing it badly. 2010-12-03T12:53:08 because of my move scheduling i had redirection too 2010-12-03T12:53:27 a1k0n: this would be move generation only. It would be evaluated by the full attack based eval so I think picking a sensible one is easy. 2010-12-03T12:53:29 so if i had a scheduled fleet and could sent it closer to the enemy while making the first intended trip in time, i would 2010-12-03T12:54:00 mega1: ah, i see. so you don't actually assume the opponent makes the move, you just find responses to potential moves and then evaluate them. 2010-12-03T12:54:01 in other words, I would not use the opponent moves when evaluating 2010-12-03T12:54:07 gotcha 2010-12-03T12:54:07 yes 2010-12-03T12:54:19 but that's still costly 2010-12-03T12:54:28 maybe not prohibitive though 2010-12-03T12:54:56 i gotta get to work, bbl 2010-12-03T12:55:07 me too 2010-12-03T13:07:16 mega1: how old are you if i may ask? 2010-12-03T13:07:22 36 2010-12-03T13:07:38 ah ok :) did you study AI ? 2010-12-03T13:07:45 oh yay, more activity 2010-12-03T13:07:52 i was affraid it would die again 2010-12-03T13:07:55 but I do look much younger :-) 2010-12-03T13:08:00 :) 2010-12-03T13:08:00 haha =) 2010-12-03T13:08:01 what's the topic? 2010-12-03T13:08:06 my age? 2010-12-03T13:08:06 like 35? 2010-12-03T13:08:09 *** amstan_ is now known as amstan 2010-12-03T13:08:17 mega1: you're like 27 2010-12-03T13:08:22 mega1: am i right? 2010-12-03T13:08:26 oh.. wait 2010-12-03T13:08:28 36 2010-12-03T13:09:00 Error323: what they taught us at uni was very old school 2010-12-03T13:09:15 he looks like 27 =) 2010-12-03T13:09:24 mega1: ah ok, but it was AI? 2010-12-03T13:09:28 it felt even less AI then machine learning does now 2010-12-03T13:09:35 Error323: they called it AI 2010-12-03T13:09:50 wlel the field hasn't changed that much? 2010-12-03T13:09:52 * Naktibalda doesn't remember what he was taught at uni, it was a crap anyway :) 2010-12-03T13:10:19 I think probability theory is starting to dominate and that's good. 2010-12-03T13:10:20 * Naktibalda spent his time in computer class playing quakeworld and counterstrike 2010-12-03T13:11:16 it makes a lot more sense then "frames" and hackish expert systems 2010-12-03T13:11:31 mega1: I agree completely 2010-12-03T13:11:37 what are frames? 2010-12-03T13:11:41 I actually found uni very very usefull 2010-12-03T13:11:54 Naktibalda: boxes with rounded corners :-) 2010-12-03T13:12:06 like UML? 2010-12-03T13:12:13 supposed to hold other frames and somehow be a good representation of knowledge. 2010-12-03T13:12:46 Naktibalda: yes, and they make me feel the same 2010-12-03T13:13:07 "Is there anything to this except for nice pictures?" 2010-12-03T13:13:24 it certainly didn't look like something that would scale. 2010-12-03T13:13:38 probability theory is fun. 2010-12-03T13:13:57 already at work? 2010-12-03T13:14:00 yes, it is 2010-12-03T13:14:01 i was felt those expert systems were dirty hacks 2010-12-03T13:14:09 i live 2 miles away 2010-12-03T13:14:31 i kinda even loathe SVMs that are so in vogue just for their geometric instead of probabilistic motivation 2010-12-03T13:14:41 same here 2010-12-03T13:14:43 i have to reframe things in terms of loss functions to talk to my coworkers 2010-12-03T13:15:28 black sheep 2010-12-03T13:16:26 there is a good critique by Hinton sayings that SVM are just flat template matchers 2010-12-03T13:17:06 and he's of course a big believer in deep architectures 2010-12-03T13:17:26 (as in several connected layers ... of neurons for example) 2010-12-03T13:17:49 hey guys something related 2010-12-03T13:18:07 how do you feel about these prediction of the singularity of ai being achieved in the year 2050 2010-12-03T13:18:13 been reading that a lot lately 2010-12-03T13:18:18 even in the scientific american 2010-12-03T13:18:40 I can't reliably judge media phenomenoms ;-) 2010-12-03T13:19:17 No I mean, its never a true researcher in the field making these claims :P 2010-12-03T13:19:29 And with good reason, imo its bullshit 2010-12-03T13:19:38 extrapolation is hard. The only one I know who's capable of it is Hari Seldon. 2010-12-03T13:19:50 there needs to be something atleast as revolutionary as the computer itself for this to happen 2010-12-03T13:20:04 stuff like reinforcement learning doesn't scale well at all 2010-12-03T13:20:28 Error323: what do you mean by reinforcement? 2010-12-03T13:20:40 learning through reward/punishment 2010-12-03T13:20:41 Make a couple of additions to the kill ant robot idea... you should take a look and let me know what you think - http://ai-contest.com/forum/viewtopic.php?f=21&t=1205 2010-12-03T13:20:45 its a field of AI 2010-12-03T13:21:47 a field of machinelearning actually, the unsupervised kind 2010-12-03T13:22:28 Perhaps reinforcement learning is fine as a top-level mechanism. 2010-12-03T13:22:37 *** fgump has joined #aichallenge 2010-12-03T13:22:46 but it needs a good unsupervised method beneath it 2010-12-03T13:22:47 I'm a big "believer" in that field 2010-12-03T13:23:32 *** medrimonia has joined #aichallenge 2010-12-03T13:23:52 it's just that using that or sth like neural nets for a singularity :P Like "all we need is computer power" is... well non-technical ppl talking P: 2010-12-03T13:23:56 *** p4p4 has quit IRC (Ping timeout: 245 seconds) 2010-12-03T13:24:52 but reading that stuff everywhere 2010-12-03T13:25:03 I wondered if I was alone in this o-O 2010-12-03T13:25:40 *** jaspervdj has quit IRC (Quit: NEVER GONNA GIVE YOU UP NEVER GONNA LET YOU DOWN) 2010-12-03T13:27:33 lol 2010-12-03T13:29:10 i've been told learn measure theory before trying to learn probability theory properly 2010-12-03T13:30:58 Error323: if the singularity happened and they had functioning bodies, we'd be fucked 2010-12-03T13:31:03 there'd be no use for us anymore 2010-12-03T13:31:17 although i'd be down with uplaoding my consciousness to a computer if possible 2010-12-03T13:31:27 preferably with a huge robot body and back ups 2010-12-03T13:31:46 probably towars the expiry date of my natural body 2010-12-03T13:32:18 I'd go with a solar powered, self-repairing space and GTFO before everything goes down the tubes here 2010-12-03T13:32:29 space ship* 2010-12-03T13:32:48 alone? :P 2010-12-03T13:33:18 Meh... after I get rid of all these hormones I think I'd be alright by myself. 2010-12-03T13:33:29 hahhaha 2010-12-03T13:33:32 nah, communication is also good 2010-12-03T13:33:48 i'd rather go down the tubes with the rest proly :P 2010-12-03T13:34:00 * Error323 non rational agent 2010-12-03T13:34:05 xD 2010-12-03T13:34:19 * antimatroid doesn't think "the singularity" is happening anytime soon 2010-12-03T13:34:23 that includes before 2050 2010-12-03T13:34:42 well I don't know, but its just impossible to predict 2010-12-03T13:34:51 i've never come across anything thorugh the ai fild anything reselmbing that kind of thing 2010-12-03T13:35:00 yet somehow the magical year 2050 solves all our problems :P 2010-12-03T13:35:04 bah, my spelling is terrible at 5 30am :P 2010-12-03T13:35:35 all ai stuff is more just how to make good approximations 2010-12-03T13:35:41 at least as far as i've looked 2010-12-03T13:36:14 (which isn't very far) 2010-12-03T13:39:03 it's the new year 2000 2010-12-03T13:39:19 indd 2010-12-03T13:39:32 numenta recently released their cortical algorithm whitepaper, which is worth looking at 2010-12-03T13:39:49 cortical? 2010-12-03T13:39:58 they don't have any kind of demo showing that it works at all 2010-12-03T13:40:07 yeah, it's based on how the neocortex is supposed to work 2010-12-03T13:40:21 http://www.numenta.com/htm-overview/htm-algorithms.php 2010-12-03T13:40:39 anyway, hinton's deep RBM is the most impressive demo i've ever seen of a neural network 2010-12-03T13:40:55 so i think there is some promise in that area 2010-12-03T13:41:07 yeah, I invested a lot of time there 2010-12-03T13:41:22 i also wrote my own rbm but haven't toyed with it all that much 2010-12-03T13:41:33 rbm being? 2010-12-03T13:41:42 restricted boltzmann machine 2010-12-03T13:41:46 i want to make a convolutional RBM like yann lecun's stuff 2010-12-03T13:42:05 that I have not implemented 2010-12-03T13:43:03 in the meantime, logistic regression on giant amounts of data works pretty well. 2010-12-03T13:43:16 (for.. non-vision problems that i am faced with) 2010-12-03T13:43:35 structure vs noise 2010-12-03T13:43:49 picture have lot of structure 2010-12-03T13:44:01 yeah, the key is getting a sparse representation somehow 2010-12-03T13:44:01 web data has little structure and lots of noise 2010-12-03T13:44:42 i wonder how hard it'd be to come up with a generative model of email 2010-12-03T13:44:52 very? 2010-12-03T13:45:30 email isn't easier then any other text, is it? 2010-12-03T13:45:43 i guess it's equivalent to the turing test at some point 2010-12-03T13:45:57 but i mainly want "email that looks like real messages" vs. "email that looks like spam" 2010-12-03T13:45:57 are you dealing with text mostly? 2010-12-03T13:46:08 text and abstract featurs of senders and recipients 2010-12-03T13:46:51 (there are temporal boltzman machine variants too) 2010-12-03T13:47:05 (and those should be a better fit for text). 2010-12-03T13:47:32 does your company actually commit money to the stock market based on its news sentiment analysis? 2010-12-03T13:47:44 I don't work there anymore. 2010-12-03T13:47:49 but they used to 2010-12-03T13:47:52 orly 2010-12-03T13:47:57 did you get any job offers yet? 2010-12-03T13:48:02 *** p4p4 has joined #aichallenge 2010-12-03T13:48:11 yes, from facebook 2010-12-03T13:48:21 more like an invitation though 2010-12-03T13:48:22 google, facebook, and startups are getting desperate here in the valley 2010-12-03T13:48:57 good "resources" are hard to come by? 2010-12-03T13:49:01 yeah. i got an invite too, and got turned down during the third phone interview cuz the guy asked me to add a special case to my code and i didn't understand what the problem was because what he said was already handled 2010-12-03T13:49:08 seems that way, yes. 2010-12-03T13:49:41 *** Migi32 has quit IRC (Quit: www.ai-contest.com) 2010-12-03T13:49:44 it was my like 5th phone interview in two weeks and i was just kinda annoyed. oh well. 2010-12-03T13:50:25 I work for Franz now and I'm quite happy with it. 2010-12-03T13:50:31 oh really? rad. 2010-12-03T13:50:49 i assume you have no plans to leave Hungary with your family and all 2010-12-03T13:50:58 *** bhasker has joined #aichallenge 2010-12-03T13:51:04 (not that i know... anything about you) 2010-12-03T13:51:15 ... but it's a consulting agreement so the future is not entirely clear. 2010-12-03T13:51:20 ah. 2010-12-03T13:51:24 yeah, you guessed right. 2010-12-03T13:51:36 We came back from Spain two years or so ago. 2010-12-03T13:51:54 ah 2010-12-03T13:52:05 so I'm happy to be a remote worker. 2010-12-03T13:52:11 it does rule. 2010-12-03T13:52:27 i kinda miss it. although having coworkers to hang out with is kinda nice too. 2010-12-03T13:52:52 I do visit every couple of months so I get to understand that my social skills are diminishing rapidly :-) 2010-12-03T13:52:57 haha 2010-12-03T13:53:35 so do you deal mostly with text nowadays? 2010-12-03T13:53:54 (or was that answered?) 2010-12-03T13:54:08 text and abstract featurs of senders and recipients 2010-12-03T13:54:12 it was 2010-12-03T13:54:17 I thought the lines crossed 2010-12-03T13:54:20 heh 2010-12-03T13:54:51 you think a TRBM could model text generatively? 2010-12-03T13:55:19 better than a non-T variant :-) 2010-12-03T13:55:38 I don't think we are at the point where anything can do a good job of it. 2010-12-03T13:56:03 yeah. bag-of-bigrams is usually good enough 2010-12-03T13:57:00 I used svm/rbm/dbn for text classification with n-grams and had slightly better results for the boltzman ones 2010-12-03T13:57:11 s/for/with/ 2010-12-03T13:57:20 interesting 2010-12-03T13:57:32 but rbm is a pain to train in comparison 2010-12-03T13:57:36 but that might have been because I liked them more and spent more time on them 2010-12-03T13:57:42 oh, yes 2010-12-03T13:57:43 hehe 2010-12-03T13:58:15 some SVMs are a pain to train too with much data 2010-12-03T13:58:23 for sentiment analysis... did you not use LDA? 2010-12-03T13:58:41 LSA 2010-12-03T13:58:50 oh, hm 2010-12-03T13:59:20 a matrix factorization model, i see 2010-12-03T13:59:34 If I recall correctly dimensionality reduction was with LSA/RBM 2010-12-03T13:59:40 then I had a classifier on top. 2010-12-03T13:59:52 SVM or another RBM. 2010-12-03T14:00:09 so did you have a lot of unclassified data you trained the RBM on or something? 2010-12-03T14:00:21 yes 2010-12-03T14:00:48 labeled data I had to fight for or worse, label it myself 2010-12-03T14:00:55 yeah, i would imagine 2010-12-03T14:01:05 *** medrimonia1 has joined #aichallenge 2010-12-03T14:01:07 *** medrimonia has quit IRC (Ping timeout: 255 seconds) 2010-12-03T14:01:28 did you have to actually come up with a vector of emotions or something? 2010-12-03T14:01:44 (or there was a convenient parallel dataset from which the learning signal was extracted) 2010-12-03T14:02:04 just a scale 2010-12-03T14:02:17 between very negative and very positive 2010-12-03T14:02:21 ah 2010-12-03T14:02:32 depends on what you want to do with it 2010-12-03T14:04:17 i dunno if you've seen http://kaggle.com/ but there's a huge variety of machine learning programming contests there 2010-12-03T14:05:05 oh dang, a $25k one ends tomorrow 2010-12-03T14:05:10 (speaking of matrix factorization, I recommend 'A Bayesian matrix Factorization Model for 2010-12-03T14:05:10 Relational Data') 2010-12-03T14:05:30 I know kaggle, but don't want to. 2010-12-03T14:05:43 i heart matrix factorization. is that what you did in the netflix prize challenge? 2010-12-03T14:05:57 i came up with it by myself and was all proud of it until i learned everyone else did too 2010-12-03T14:07:06 cool. I don't think I did too many original things there. 2010-12-03T14:07:19 but got the boltzmann bug. 2010-12-03T14:07:51 I wanted to get some kind of ant system working. 2010-12-03T14:08:05 without any success whatsoever. 2010-12-03T14:10:48 *** Migi32 has joined #aichallenge 2010-12-03T14:10:50 haha i was thinking of that too 2010-12-03T14:11:07 i'm working my way through 'on intelligence' by jeff hawkins 2010-12-03T14:11:30 is it good? 2010-12-03T14:11:32 as i said there's no public proof that his cortex ideas actually work but the rbm stuff is similar enough to what he describes and does seem to work 2010-12-03T14:11:37 it's very interesting. yeah, recommended. 2010-12-03T14:11:56 *** Migi32 has quit IRC (Client Quit) 2010-12-03T14:11:59 although a little opaque. 'layer 4 sends a feedforward signal up the column to layers 3 and 2 and then layer 1 sends to layer 6' 2010-12-03T14:12:05 *** Migi32 has joined #aichallenge 2010-12-03T14:12:25 i'm having trouble following the details 2010-12-03T14:13:06 I have never heard of a boltzmann machine o_O 2010-12-03T14:13:22 yet it seems quite significant for an AI student lol 2010-12-03T14:13:28 Error323: http://www.youtube.com/watch?v=AyzOUbkUf3M 2010-12-03T14:13:59 a1k0n: thanks 2010-12-03T14:14:07 is this very new? 2010-12-03T14:14:13 a couple years old now 2010-12-03T14:14:22 Error323: I also have a couple of blog entries on them. 2010-12-03T14:14:35 ... although focusing on my library. 2010-12-03T14:14:43 i mean, boltzmann machines have been around for a long time but hinton's learning algorithm seems to be a sort of breakthrough 2010-12-03T14:14:47 (recent breakthrough) 2010-12-03T14:14:57 cool 2010-12-03T14:15:01 ill delve into it 2010-12-03T14:15:34 a1k0n: are aware of this: http://www.kyb.mpg.de/bs/people/weston/papers/unified_nlp.pdf ? 2010-12-03T14:15:46 "A Unified Architecture for Natural Language Processing: Deep Neural Networks with Multitask Learning" 2010-12-03T14:15:53 *are you 2010-12-03T14:16:05 i don't think i had seen that, no. thanks. 2010-12-03T14:16:10 it's great 2010-12-03T14:16:34 there are signs that rbm are not the only deep learners that work and this is one 2010-12-03T14:17:17 nice. yeah, lecun has a couple non-rbm deep networks, usually alternating convolutional/non-convolutional layers 2010-12-03T14:17:49 yes, that's his thing 2010-12-03T14:18:13 now is there something like convolution for non-vision tasks? 2010-12-03T14:18:30 the paper you just linked me uses convolutional weight sharing 2010-12-03T14:18:44 heh, I forgot 2010-12-03T14:19:27 you will find the coveted distributed representation there as well if I'm not mistaken 2010-12-03T14:23:10 *** janzert1 has joined #aichallenge 2010-12-03T14:24:49 LOl the guy is quite funny :D 2010-12-03T14:24:55 cool techtalk 2010-12-03T14:24:56 *** janzert has quit IRC (Ping timeout: 250 seconds) 2010-12-03T14:30:54 damn this is awesome 2010-12-03T14:34:18 *** Naktibalda has quit IRC (Quit: ChatZilla 0.9.86 [Firefox 3.6.8/20100722155716]) 2010-12-03T14:44:08 *** contestbot_ has joined #aichallenge 2010-12-03T14:44:12 *** contestbot has quit IRC (Disconnected by services) 2010-12-03T14:44:19 *** amstan has quit IRC (Ping timeout: 255 seconds) 2010-12-03T14:46:04 *** contestbot_ is now known as contestbot 2010-12-03T14:46:05 *** ChanServ sets mode: +o contestbot 2010-12-03T14:48:05 *** RainCT has quit IRC (Quit: cya) 2010-12-03T14:50:11 *** amstan_ is now known as amstan 2010-12-03T14:51:31 *** tapwater has joined #aichallenge 2010-12-03T14:54:09 *** Appleman1234 has quit IRC (Ping timeout: 276 seconds) 2010-12-03T15:00:23 *** Naktibalda has joined #aichallenge 2010-12-03T15:07:30 FUUUU youtube 2010-12-03T15:07:46 no i didn't want you to reload the entire 50 minute video just cause i exited full screen 2010-12-03T15:07:48 ffs! 2010-12-03T15:07:58 could have been chromes fault 2010-12-03T15:08:01 same damn company! 2010-12-03T15:08:37 lol 2010-12-03T15:09:12 hahaha 2010-12-03T15:11:12 *** fgump has quit IRC (Quit: Ex-Chat) 2010-12-03T15:21:23 *** fgump has joined #aichallenge 2010-12-03T15:22:58 *** boegel has joined #aichallenge 2010-12-03T15:30:52 *** Cold-Phoenix has quit IRC () 2010-12-03T15:34:14 *** lavalamp has joined #aichallenge 2010-12-03T15:37:41 *** RainCT has joined #aichallenge 2010-12-03T15:40:47 *** RainCT has quit IRC (Remote host closed the connection) 2010-12-03T15:41:51 *** narnach has quit IRC (Ping timeout: 240 seconds) 2010-12-03T15:44:17 *** sigh has joined #aichallenge 2010-12-03T15:47:24 *** narnach has joined #aichallenge 2010-12-03T15:49:03 *** flag has quit IRC (Quit: Lost terminal) 2010-12-03T15:54:33 *** narnach has quit IRC (Quit: Leaving.) 2010-12-03T15:55:42 *** bhasker_ has joined #aichallenge 2010-12-03T15:57:43 *** yasith has quit IRC (Read error: Operation timed out) 2010-12-03T15:57:59 *** bhasker has quit IRC (Read error: Operation timed out) 2010-12-03T15:59:26 *** bhasker has joined #aichallenge 2010-12-03T15:59:49 *** phreeza has quit IRC (Read error: Connection reset by peer) 2010-12-03T16:00:16 *** phreeza has joined #aichallenge 2010-12-03T16:02:19 *** bhasker_ has quit IRC (Ping timeout: 272 seconds) 2010-12-03T16:06:12 *** phreeza has quit IRC (Quit: Computer has gone to sleep) 2010-12-03T16:11:07 *** yasith has joined #aichallenge 2010-12-03T16:14:50 *** Appleman1234 has joined #aichallenge 2010-12-03T16:16:46 *** p4p4 has quit IRC (Quit: ChatZilla 0.9.84 [SeaMonkey 2.0a3/20090223135443]) 2010-12-03T16:21:00 *** Appleman1234 has quit IRC (Ping timeout: 265 seconds) 2010-12-03T16:26:34 *** gump_ has joined #aichallenge 2010-12-03T16:26:56 *** fgump has quit IRC (Ping timeout: 240 seconds) 2010-12-03T16:27:46 hmm 2010-12-03T16:31:55 *** Mathnerd314 has joined #aichallenge 2010-12-03T16:32:22 *** Rubicon-|-Cross has joined #aichallenge 2010-12-03T16:32:37 *** boegel has quit IRC (Quit: Leaving) 2010-12-03T16:38:53 *** Appleman1234 has joined #aichallenge 2010-12-03T16:41:13 *** Appleman1234 has quit IRC (Excess Flood) 2010-12-03T16:41:26 *** Mathnerd314 has quit IRC (Ping timeout: 245 seconds) 2010-12-03T16:43:25 *** bhasker_ has joined #aichallenge 2010-12-03T16:45:57 *** bhasker has quit IRC (Ping timeout: 276 seconds) 2010-12-03T16:45:57 *** wh1teside_ has quit IRC (Read error: Connection reset by peer) 2010-12-03T16:46:16 *** wh1teside has joined #aichallenge 2010-12-03T16:50:41 *** Appleman1234 has joined #aichallenge 2010-12-03T16:51:09 *** sigh has quit IRC (Remote host closed the connection) 2010-12-03T17:04:14 *** amstan has quit IRC (Ping timeout: 265 seconds) 2010-12-03T17:06:26 *** mega1 has quit IRC (Ping timeout: 245 seconds) 2010-12-03T17:11:58 *** dmj111 has joined #aichallenge 2010-12-03T17:25:29 *** jaspervdj has joined #aichallenge 2010-12-03T17:41:57 *** sigh has joined #aichallenge 2010-12-03T17:43:00 hmm indeed 2010-12-03T17:44:20 hmmm too 2010-12-03T17:44:39 *** medrimonia1 is now known as Medrimonia 2010-12-03T17:49:52 wtf... why do so many articles about this contest focus on the languages used, and use the contest results to make ridiculous assertion 2010-12-03T17:50:01 :/ 2010-12-03T17:50:53 Zannick: did you know that Lisp is the greatest language for AI, and all other languages don't even come close? 2010-12-03T17:50:55 The fact that the rankings are a wild mess of languages pretty much indicates that you can't say one language is strictly better than another 2010-12-03T17:51:51 yup 2010-12-03T17:54:02 saying that lisp is the best language for ai is saying that lisp was the language of the winning entry in the ai challenge this time around, and that the ai contest is such a good litmus test of real ai that the language of the top entry must be the best language for ai 2010-12-03T17:54:15 and only one of those assertions holds 2010-12-03T17:55:22 yeah 2010-12-03T17:57:27 *** Appleman1234 has quit IRC (Ping timeout: 240 seconds) 2010-12-03T18:01:53 *** ChanServ sets mode: +o Zannick 2010-12-03T18:02:08 *** Zannick sets mode: -b zz_SophieB!*@* 2010-12-03T18:02:19 *** Zannick sets mode: -b *!*@213.88.2.78 2010-12-03T18:02:23 *** Zannick sets mode: -b *!*@*213.88.2.78 2010-12-03T18:02:36 *** Zannick sets mode: -o Zannick 2010-12-03T18:02:48 *** ChanServ sets mode: +o Zannick 2010-12-03T18:02:56 *** Zannick sets mode: -b borderland!*@* 2010-12-03T18:03:05 *** Zannick sets mode: -o Zannick 2010-12-03T18:13:01 *** bhasker has joined #aichallenge 2010-12-03T18:15:50 *** bhasker_ has quit IRC (Ping timeout: 260 seconds) 2010-12-03T18:21:22 *** Azrathud has joined #aichallenge 2010-12-03T18:32:32 clearly it's the combination of language, country of residence, and age.. 2010-12-03T18:32:40 one should move, and wait till a certain age before using lisp 2010-12-03T18:33:14 *** lavalamp has quit IRC (Quit: Page closed) 2010-12-03T18:39:01 *** Erika_ has joined #aichallenge 2010-12-03T18:39:56 haha 2010-12-03T18:41:40 *** Appleman1234 has joined #aichallenge 2010-12-03T18:41:41 next year janzert should choose the maps so that a 14 yo PHP developer from canada win, instead of selecting some random Lisp with a background in AI won 2010-12-03T18:42:21 s/won/to win 2010-12-03T18:42:37 the winning goal in a soccer game was scored by someone with a broken leg..so I'm gona jump off a building so I can becoem a pro soccer player 2010-12-03T18:43:49 *** Erika_ has quit IRC (Ping timeout: 272 seconds) 2010-12-03T18:46:54 shark attacks and ice cream cause summer! 2010-12-03T18:48:10 i am going to rewrite my bot in a neater way now that the pressure is off 2010-12-03T18:48:17 so i can tinker at peace 2010-12-03T18:48:45 until it can beat boscimacko? :P 2010-12-03T18:51:44 maybe maybe not 2010-12-03T18:51:54 but at least i want to implement a proper minmax solution or something 2010-12-03T18:52:09 so that at least i know how the whole thing is done 2010-12-03T18:52:25 what do you mean? 2010-12-03T18:52:39 well i have never really implemented a minxmax solution 2010-12-03T18:52:47 so just to get a hang of how to explore the game tree etc 2010-12-03T18:52:49 ah ok, fair enough 2010-12-03T18:53:02 and how to implement basic alpha beta pruning 2010-12-03T18:53:09 etc etc 2010-12-03T18:53:19 good luck, learning minimax using planetwars is a baptism of fire :P 2010-12-03T18:53:37 well i will just simulate it like a turn by turn 2010-12-03T18:53:46 make it non simultaneous 2010-12-03T18:53:47 for now 2010-12-03T18:53:57 assume that i move first and then the enemy 2010-12-03T18:54:09 yeah 2010-12-03T18:54:15 i looked at the codecup competition 2010-12-03T18:54:20 but the most important thing will be move selection 2010-12-03T18:54:21 but dvonn doesn't seem interesting 2010-12-03T18:54:31 i think i have a decent move generator 2010-12-03T18:54:35 and an evaluation function 2010-12-03T18:54:52 so it should be good enough to start off 2010-12-03T18:55:31 :) 2010-12-03T18:56:16 be sure to test it against other bots though, running minimax against yourself is not going to be a god indicator of how well it does :P 2010-12-03T18:56:21 given the massive pruning 2010-12-03T18:56:41 but there are plenty of bots available so, that should not be a problem 2010-12-03T18:57:28 yea and my heuristic bot should do fine 2010-12-03T18:57:37 since it ended up 54 so it won't be a bad test bed 2010-12-03T18:58:06 i think initially i am just going to generate one set of moves per planet 2010-12-03T18:58:15 no, my point was that it will be a bad test bed if you are using the same heuristics for move generation in minimax 2010-12-03T18:59:18 ah k 2010-12-03T18:59:40 did you read about codecup 2010-12-03T18:59:50 they have much stricter time requirements for the game 2010-12-03T18:59:55 5 seconds for the whole game ^ 2010-12-03T18:59:59 I glanced at it when antimatroid posted about it 2010-12-03T19:00:09 though watching the game didnt look very interesting 2010-12-03T19:00:17 didn't strike me as an interesting game 2010-12-03T19:00:18 yeah 2010-12-03T19:01:19 oh they do have an applet visualizer 2010-12-03T19:04:00 yoooooooooooooooooooooo 2010-12-03T19:04:10 o_O 2010-12-03T19:04:17 :P 2010-12-03T19:04:19 hi 2010-12-03T19:04:20 i mena 2010-12-03T19:04:22 mean* 2010-12-03T19:04:29 :P 2010-12-03T19:05:01 ^_^ 2010-12-03T19:05:26 cool, so I start talking 2010-12-03T19:05:29 and everyone stops 2010-12-03T19:05:32 *** bhasker_ has joined #aichallenge 2010-12-03T19:05:34 xD 2010-12-03T19:07:13 *** bhasker has quit IRC (Ping timeout: 264 seconds) 2010-12-03T19:09:31 *** Mathnerd314 has joined #aichallenge 2010-12-03T19:23:45 *** mceier has quit IRC (Quit: leaving) 2010-12-03T19:47:10 hey, who had the infinite map in javascript? 2010-12-03T19:52:18 *** Cyndre has left #aichallenge ("Leaving") 2010-12-03T19:55:47 everyone recover yet? today i didn't accidentally check my latest games on the webpage, unlike yesterday. 2010-12-03T19:56:03 hehe its kinda sad to see the rankings frozen 2010-12-03T19:56:09 though its good to have a life back 2010-12-03T19:56:13 :) 2010-12-03T19:57:07 he, that is for sure. except it feels a little weird still. 2010-12-03T19:57:14 *** bhasker_ has quit IRC (Quit: bhasker_) 2010-12-03T20:05:34 *** Cyndre has joined #aichallenge 2010-12-03T20:07:25 *** yasith has quit IRC (Ping timeout: 272 seconds) 2010-12-03T20:15:54 *** Meatkat has quit IRC (Ping timeout: 276 seconds) 2010-12-03T20:19:51 *** yasith has joined #aichallenge 2010-12-03T20:20:19 *** Appleman1234 has quit IRC (Remote host closed the connection) 2010-12-03T20:47:03 *** virdo has joined #aichallenge 2010-12-03T20:51:44 *** gump_ has quit IRC (Quit: Ex-Chat) 2010-12-03T21:11:16 *** amstan has joined #aichallenge 2010-12-03T21:11:16 *** ChanServ sets mode: +o amstan 2010-12-03T21:15:01 *** wh1teside has quit IRC (Ping timeout: 255 seconds) 2010-12-03T21:16:15 hahaha 2010-12-03T21:16:27 i just realized galcon is short for galaxy conquest 2010-12-03T21:16:50 *** wh1teside has joined #aichallenge 2010-12-03T21:16:50 i've been pronouncing it gal-son 2010-12-03T21:16:58 lol 2010-12-03T21:30:32 *** jonoff has quit IRC (Ping timeout: 265 seconds) 2010-12-03T21:35:33 I've been pronouncing it glay-ken 2010-12-03T21:35:50 er.. glay-kin 2010-12-03T21:36:28 *** yasith has quit IRC (Ping timeout: 245 seconds) 2010-12-03T21:36:57 I'm either dislexic or retarded 2010-12-03T21:36:59 or both :) 2010-12-03T21:39:57 haha. allo! galçon! 2010-12-03T21:40:19 a1k0n: that is actually what i was thinking of. 2010-12-03T21:40:33 just didn't feel like getting the special character 2010-12-03T21:40:49 protip: screen digraphs 2010-12-03T21:41:11 i'm using screen...it does that? 2010-12-03T21:41:18 ^A^Vc, 2010-12-03T21:41:23 don't you know, screen does everything 2010-12-03T21:41:30 whoa 2010-12-03T21:41:44 évèrythîng 2010-12-03T21:41:55 great for leetspeak. 2010-12-03T21:42:03 cræzy 2010-12-03T21:43:10 *** sigh has quit IRC (Remote host closed the connection) 2010-12-03T21:56:15 *** Appleman1234 has joined #aichallenge 2010-12-03T22:02:22 Could I open Gmail in another email client? 2010-12-03T22:02:38 Er, my email address in another email client. 2010-12-03T22:02:49 my email address' account*, rather 2010-12-03T22:04:48 screen is pretty amazing 2010-12-03T22:04:57 other useful features: ^A^G for bell toggle 2010-12-03T22:05:04 ^A^C for instant shell 2010-12-03T22:05:21 great for irc :] 2010-12-03T22:15:47 *** Naktibalda has quit IRC (Remote host closed the connection) 2010-12-03T22:24:42 *** Migi32 has quit IRC (Quit: BOINC) 2010-12-03T22:29:49 *** Mjothvitnir has joined #aichallenge 2010-12-03T22:45:18 *** Mjothvitnir has quit IRC (Quit: Page closed) 2010-12-03T23:25:35 *** amriedle has joined #aichallenge 2010-12-03T23:27:41 moo 2010-12-03T23:30:23 yes! cow, someone get the hide, i need that armor 2010-12-03T23:40:15 wat] 2010-12-03T23:50:24 *** amriedle has quit IRC (Quit: Lost terminal)