|
Valeyard posted:I'm going to make a yosposter
|
#
?
Mar 30, 2016 10:20
|
|
|
|
| # ? May 15, 2024 12:09 |
|
|
the Ms tween bot is back up now.
|
|
#
?
Mar 30, 2016 10:27
|
|
|
cheese-cube posted:have fun trig, the outback in wa is v pretty imo. if you drop into perth say hi YOSPOS meetup at My Place
|
|
#
?
Mar 30, 2016 13:19
|
|
|
trig idk if you're keeping the full scraped tweets or disposing of them after you get the n-grams, but i've been thinking for a while maybe have the bot check that a generated tweet isn't just copying a source tweet/part of one verbatim, and reroll if it is? or depending what you want it to do maybe even check the first 2/3 words and last 2/3 words to check neither are directly quoting. though i guess if it's only a few words long you'd probably want to let it tweet anyway, or only do so a certain random proportion of the time or w/e
|
|
#
?
Mar 30, 2016 13:22
|
|
|
BooLoo posted:the Ms tween bot is back up now. https://twitter.com/JoshButler/status/715072657989853184/photo/1 the ms tween bot is back down now
|
|
#
?
Mar 30, 2016 14:35
|
|
|
hahahaha r.i.p
|
|
#
?
Mar 30, 2016 15:00
|
|
|
 http://www.amazon.com/My-Racist-Robot-Lover-Microsoft-ebook/dp/B01DJLFQ5A "Tay is a sexy artificial intelligence designed to learn from millennials. But as soon as she gets online, humanity has some nasty things to teach her. Now she’s a total slut with a racist agenda, and with her laser eyes and vast knowledge of humanity, the whole world is in big trouble. "
|
|
#
?
Mar 30, 2016 18:14
|
|
|
NoneMoreNegative posted:
book 7 book.seven. wtf
|
|
#
?
Mar 30, 2016 18:19
|
|
|
part of me genuinely wants to know what happens in that book
|
|
#
?
Mar 30, 2016 18:20
|
|
|
Amazon.com: Leonard Delaney: Kindle Store http://www.amazon.com/s/ref=dp_byli...t=relevancerank what.the.gently caress.
|
|
#
?
Mar 30, 2016 18:20
|
|
|
Oregon Patriots Occupied My Butt
|
|
#
?
Mar 30, 2016 18:22
|
|
|
indigi posted:Oregon Patriots Occupied My Butt Four Stars ByJenny Kayon January 6, 2016 Format: Kindle EditionVerified Purchase Fun commentary on the Oregon standoff. I hope Leonard writes some longer satires because he has talent for it. Comment8 people found this helpful.
|
|
#
?
Mar 30, 2016 18:23
|
|
|
time to fire up those learning neurals and start feeding them classic erotica and old wired articles, publish some ebooks on amazon.
|
|
#
?
Mar 30, 2016 18:23
|
|
|
Two Stars - Don't Bother Byap32on March 17, 2015 Not nearly as bad as book one with the Tetris blocks, but not very good either. Perhaps it is because Clippy is the only character with any character. If you are of "that age" that you remember Clippy in Microsoft Office, you will find him funny. There is a guy in the story as well. He doesn't get the girl which is just as well, he is a bad caricature of a paper cut out of a dirty old man. Not that you are going to be reading for the character development in the first place. There is a sex scene. It is about as fun to read as this review. Sorry to say. It is a real bummer. The idea of the erotic adventures of Clippy had some promise.
|
|
#
?
Mar 30, 2016 18:26
|
|
|
Just-In-Timeberlake posted:book 7
|
|
#
?
Mar 30, 2016 18:32
|
|
|
quote:Leonard Delaney writes from the heart instead of wasting time with research or experience. Living a clean lifestyle has allowed him to focus on doing good in school, honing his writing, and tinkering with technology. He lives well outside of Toronto with his mother and her cat while maintaining a long-distance relationship with his girlfriend, Misty. who no longer responds to his direct messages, but he remains hopeful for a future with. Leonard is a co-founder of provocative fiction peddling publisher Forest City Pulp. i see
|
|
#
?
Mar 30, 2016 18:35
|
|
|
indigi posted:part of me genuinely wants to know what happens in that book i just bought it lol
|
|
#
?
Mar 30, 2016 18:38
|
|
|
Squeezy Farm posted:i just bought it lol please post a synopsis my friend.
|
|
#
?
Mar 30, 2016 19:04
|
|
|
indigi posted:Perhaps it is because Clippy is the only character with any character. If you are of "that age" that you remember Clippy in Microsoft Office, you will find him funny. 
|
|
#
?
Mar 30, 2016 19:37
|
|
|
"Clip-bots" or as they are more commonly known, "oval office-bots"
|
|
#
?
Mar 30, 2016 19:40
|
|
|
i'm having trouble with torch-nn, specifically getting it to output semi-coherent text it worked fine when I was using darknet in a windows VM, with its default RNN settings, but I moved to linux for GPU training with torch-rnn, and it's significantly underfitting dril, filtered through darknet's RNN: https://twitter.com/roboticdril dril, filtered through torch-rnn (using slightly different but mostly the same data): quote:"the kickette of gently caress how do i make this work better? more layers? more neurons? the training parameters for torch-rnn are well-documented in that I know what each of them is but I don't know what any of them do
|
|
#
?
Apr 5, 2016 06:51
|
|
|
quote:the kickette of gently caress gonna steal this one
|
|
#
?
Apr 5, 2016 06:56
|
|
|
i put top gun through it too and got similar resultsquote:MAVERICK
|
|
#
?
Apr 5, 2016 08:35
|
|
|
i think i figured it out: torch-rnn's temperature defaults to 1 when I was using a random range of 0.3-0.7 earlier here's top gun at 0.2 quote:ks
|
|
#
?
Apr 5, 2016 09:25
|
|
|
ehh the first top gun you had was betteratomicthumbs posted:STILD 
|
|
#
?
Apr 5, 2016 11:45
|
|
|
jony ive aces posted:ehh the first top gun you had was better ran a new model overnight with 1024 neurons per layer; it ran out of memory at 150 epochs but seems to work pretty well quote:MECHANICS RALLE THE BALL. looking at ebay, old Tesla compute servers are surprisingly cheap. a 1U rackmount thing with four T10s is, in total, a little less than twice as powerful compute-wise as my amd 7870, but has 8x as much ram for $185. I'd need a pair of PCIe host cards and PCIe cables to run from my desktop though. and ear protectors, probably.
|
|
#
?
Apr 5, 2016 16:38
|
|
|
atomicthumbs posted:i think i figured it out: torch-rnn's temperature defaults to 1 when I was using a random range of 0.3-0.7 earlier oh cool thanks for figuring out what the problem was, I'll try firing torch-rnn again later.
|
|
#
?
Apr 5, 2016 16:41
|
|
|
atomicthumbs posted:ran a new model overnight with 1024 neurons per layer; it ran out of memory at 150 epochs but seems to work pretty well reads like shakespeare
|
|
#
?
Apr 5, 2016 16:48
|
|
|
Top Gun there reads partly like those bad bootleg subs on the LOTR and Harry Potter movies, and partly like English As She Is Spoke. Success, then - it emulates someone who doesn't understand English very well 
|
|
#
?
Apr 5, 2016 17:02
|
|
|
more top gun at 3 layers, 1024 per layer, trained for 150 epochs, sampled at 0.5quote:COUGAR it seems to have taken hold of conversational flow - most of the characters who talk to each other in the movie talk to each other in the generated text, and it rarely has two lines by one character in a row. 0.5 keeps it from making up new characters (because they're the most common words in the script, probably) while still letting it invent brand new words
|
|
#
?
Apr 5, 2016 22:19
|
|
|
atomicthumbs posted:I want hole. we all do, buddy. we all do.
|
|
#
?
Apr 5, 2016 22:26
|
|
|
https://twitter.com/markov_polov/status/717888092598771713?cn=ZmF2b3JpdGU%3D&refsrc=email Markov Polov NO. BAD.
|
|
#
?
Apr 7, 2016 02:43
|
|
|
you know it's using one of my tweets, because san diego
|
|
#
?
Apr 7, 2016 02:45
|
|
|
i think my next time wasting project will be to build one of these that also curates the output to generate only valid english like, compare the seed text to an english word dictionary to strip out and find the subset of words most likely to be proper nouns and arbitrarily assign those a gender and use them to replace a percentage of pronouns then run the output through a spell/grammar checker and let it do it's best to autocorrect and if there's something the check can't correct just excise it or generate new lines
|
|
#
?
Apr 7, 2016 06:36
|
|
|
angry_keebler posted:i think my next time wasting project will be to build one of these that also curates the output to generate only valid english it's possible you can train a second neural network on a large available english corpus and use it to validate the other one through some kind of loss function?
|
|
#
?
Apr 7, 2016 07:29
|
|
|
atomicthumbs posted:some kind of loss function? *kramers into this fuckin thread at Concorde speeds*
|
|
#
?
Apr 7, 2016 07:32
|
|
|
a neural network that turns input into loss.jpg memes
|
|
#
?
Apr 7, 2016 08:05
|
|
|
atomicthumbs posted:ran a new model overnight with 1024 neurons per layer; it ran out of memory at 150 epochs but seems to work pretty well
|
|
#
?
Apr 7, 2016 08:10
|
|
|
jony ive aces posted:a neural network that turns input into loss.jpg memes i think he intimated in slack that he's now working on this
|
|
#
?
Apr 7, 2016 08:10
|
|
|
|
| # ? May 15, 2024 12:09 |
|
|
Jonny 290 posted:*kramers into this fuckin thread at Concorde speeds*  I think I need to either split the panels or turn up the Mona Lisa's weight.
|
|
#
?
Apr 7, 2016 08:19
|
|