First of all, neural nets learn by example. If you show it 1,000 hours of video (assuming 120,000 unique 30-sec Olive Garden commercials exist), you’ll get video out, not a script with stage directions. 2/12
Show this thread
These "I forced a bot to watch X" posts are almost certainly 100% human-written with no bot involved. Here's how you can tell. 1/12https://twitter.com/KeatonPatti/status/1006961202998726665 …
-
-
-
Notice that this script has the same main characters and scenario the entire way through. An actual neural net’s story will tend to meander dreamlike because it forgets what it was doing. 3/12
Show this thread -
Here’s a recipe written by an actual neural net that examined 30,000 recipes. It gets the overall structure okay, but by the time it reaches the directions it has forgotten the ingredients list. http://aiweirdness.com/post/159022733587/three-bad-recipes-generated-by-neural-network … 4/12pic.twitter.com/pRZJpg9nap
Show this thread -
Neural nets also have trouble contructing complex sentences, unless its data contains lots of examples of that particular type. 5/12
Show this thread -
Here’s a few bits by a neural net that read over 100,000 Harry Potter fan fiction summaries. When the sentences get complex, it stumbles on the grammar. http://aiweirdness.com/post/164291045392/harry-potter-and-the-word-level-recurrent-neural … http://aiweirdness.com/post/162668008357/harry-potter-and-the-neural-network-fan-fiction … 6/12pic.twitter.com/a4kGusu67q
Show this thread -
It’s not *impossible* to train a neural net that can keep track of characters in a story, and write generally good grammar. But it would be clunkier than this, and you’d get major kudos for managing it at all. 7/12
Show this thread -
Also not to be confused with
@botnikstudios who train predictive keyboards, which a human uses to write scripts. That’s how they get complete sentences and coherent plots like this one. 8/12https://twitter.com/botnikstudios/status/959145510899560448 …
Show this thread -
They’ve published their keyboards; anyone can use them or train their own. For scripts they usually have separate keyboards for dialog and stage directions, and a human decides when to switch between them. http://botnik.org/apps/ 9/12pic.twitter.com/LkrZxWjoPL
Show this thread -
The “I forced a bot” script isn’t even a predictive keyboard. Pretty sure “nachos” didn’t appear in its “training data”. 10/12
Show this thread -
Even when I publish lists of neural net-generated words, there’s human involvement: I’m sorting through the output to pick the most interesting names. http://aiweirdness.com/post/166814009412/a-neural-network-designs-halloween-costumes … 11/12pic.twitter.com/HHgm9pjbXf
Show this thread -
I wish people wouldn’t present these fakes as bot-written. Actual AI-written text just isn’t that coherent. 12/12
Show this thread -
But the bit about "unlimited stick" is pretty darn funny. 13/12.
Show this thread
End of conversation
New conversation -
-
-
It's funny, I assumed that was obvious given the author is a comedy writer -- but then I realized, nobody else is looking at that, and this probably does look that way. Huh.
-
And there ARE comedy writers who use real machine learning in their work. Actual AI-generated humor should be recognized. My mentions right now are 50% people going "duh, comedy writer, i knew it was fake" and 50% people going "omg thank you i learned so much"
-
I’m in the THANK YOU department for making me realize something obvious to me as a writer isn’t obvious to the casual viewer.
-
I still talk to people who are genuinely worried that computers are almost smart enough to take over from us. Maybe if "I forced a bot" becomes a meme, it will actually help dispel this. My thread will look silly then, but oh, well.
-
YES! Thank you thank you. People don't listen to me when I try to tell them this, because I'm not considered an "AI expert" to them. I'm just a lowly sysadmin who has worked with machine learning quite a bit. Yes, very hard to dispel! "but the government!! They have stuff!"
-
have them try a few conversations with this bot. This is actually state of the art.https://twitter.com/janellecshane/status/1008930047384694784 …
End of conversation
New conversation -
-
-
Is there a bot that can tell me whether or not something was written by a bot?
-
I would love to play with one. People have tried adversarial text generation, which would necessarily involve one algorithm that's trying to tell the difference between human-generated and algorithm-generated text.
-
I tried to make a text-GAN and it wasn't pretty. All papers on text-GANs are basically conclude with a ¯\_(ツ)_/¯
-
Oh darn. Was the discriminator any fun, at least?
-
The discriminator ended up being too smart so I made it dumber. But then the generator just generated random lowercase text of the same length and that was enough to fool it.
-
If you allow the discriminator long term interactions it should be pretty easy, right? Seems they would be easier to include in the discriminator than the generator.
-
Including an RNN-layer in the discriminator architecture is exactly what made it too smart :\
-
Could the discriminator be used as standalone? Like, could you feed it *any* text and see its verdict of human vs computer? I'd be curious to see what it does with the styles of various authors.
- 1 more reply
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.