As I catch you on your way out to a food-filled vacation, here’s a party conversation-starter (and hangover cure) to keep in your back pocket.
Today I published a story on my misadventures in using neural networks to generate Christmas movie summaries—first recounted in The Algorithm on Tuesday. The tl;dr? I fed 360 movie plots from Wikipedia into an algorithm and fell into a deep rabbit hole of trying to get the funniest results. If you haven’t yet, you can read a saga of my process here. But we all know the best part is the results.
So, without further ado.
I present to you:
The best synopses
- A family of the Christmas terrorist and offering the first time to be a charlichhold for a new town to fight.
- A woman from chaos adopted home believes.
- A gardener detective but country murderer magical suddenly Christmas the near elf.
- A intercepting suffers and a friends up change Christmas with his and save Christmas time.
- A man returns to the singer who is forced to return his life with a couple to help her daughter for Christmas.
- Lonely courier village newspaper by home destroy Christmas Christmas Christmas the prancer.
- Babysitter boy tries to party the Christmas in of for more Christmas.
The best titles
- Santa Christmas Christmas
- The Christmas Mile
- Fight Christmas
- The Nighht Claus
- I Santa Manta Christmas Porie
- Babee Christmas
- A Christmas StorK
- Christmas Caper
- The Lick Christmas
- Mrack Me Christmas Satra
- The Christmas Catond 2
- It Santa
The full story also includes kick-ass movie posters, courtesy of our lovely art department, such as:
Here's the link to the story again. It features some excellently lame annotations on all of the synopses and titles. Deputy editor Michael Reilly and I went a little overboard. You’re welcome.
Toxic trolls. Social media companies like Facebook and Twitter have struggled consistently to automatically identify and stop harassment accurately and reliably on their platforms. For all the recent progress in AI, you can easily thwart an algorithm from finding abusive messages by disguising them in sarcasm or with a sprinkling of positive keywords.
A new study is now trying to change that. Amnesty International partnered with researchers at ElementAI to develop a better method for classifying toxic tweets. The researchers first collected abusive tweets with a machine-learning tool similar to the one used to classify spam. They then gave volunteers a mix of pre-classified and previously unseen tweets to classify, which became the training data for a neural network.
The resultant system was able to identify abuse with impressive accuracy and give Amnesty International a better snapshot of the scale of harassment on Twitter. It found that 1.1 million abusive tweets were sent in 2017 to the 778 female politicians and journalists they studied—equivalent to one every 30 seconds. The algorithm is now also available as a tool called Troll Patrol. Read more here.
Bits and Bytes
Facebook relies too much on AI to solve its content moderation problems
But there also isn’t a clear alternative. (WIRED)
China is surging ahead in AI compared to the US
A lack of privacy complaints, safeguards, or regulatory constraints has allowed the technology to rapidly develop. (TR)
Machine vision can create Harry Potter-style photos for muggles
A clever algorithm animates characters in still images, allowing them to walk out of photographs. (TR)
Deep learning helped to map every solar panel in the US
It was able to identify 1.47 million solar installations from satellite images, exceeding previous estimates. (TR)
A researcher made a supercomputer out of thousands of virtual game boys
It could help AI learn faster by transferring knowledge from one game to another. (Motherboard)