Profile picture
Janelle Shane
@JanelleCShane
, 4 tweets, 2 min read Read on Twitter
Fascinated by the way this crochet-trained neural net deals with non-crochet prompts.
Since HAT3000 started out as vanilla GPT-2 345M trained on huge sections on the internet, it still remembers how to generate stories and fan fiction. But now it tries its hardest to gradually turn them into crochet hat patterns. Prompt: Darth Vader clenched his fist.<br /> <br /> Darth Vader clenched his fist. “I am not the Darth,” he cried, throwing his hood up and closing his eyes tight. “I am not… you do not know me, you do not recognize me, and I am not supposed to share who I am with you… only that I know you are not supposed to make these fools of yourselves!”<br /> <br /> The Force touched him and his smile widened. “I am Alderaan,” he said, glancing at the
This one I find particularly interesting - HAT3000’s story isn’t about hats at first, but is weirdly conceptually similar. Prompt: “The earthlings did not yet know they were doomed.”<br /> <br /> The earthlings did not yet know they were doomed. But in the round more vivid did they see the earth walking, as it was moving about the west of the round, and at the same time it erected a hedge round the central part of the circle. Then as the round demanded scenes, the earthling made a ring of earth, and placed the remaining clay in place.<br /> “Now fill the
HAT3000's stories often end up with intricately choreographed fight scenes. The resemblance to crochet pattern directions is striking (even before the AI introduces a flying ball of yarn). Kylo Ren leapt behind a support column at the sight of Han Solo. He swiped at his left eye and left wrist, and with one swift movement spun on his heel for his next swing. The spinning action knocked a handful of the dust from Han Solo’s eyes and onto his feet, and soon he was spinning in the air and making his way around the platform without stopping. His arm suddenly snapped to his side like a mallet’s head snappinLoki held out his hand, laughing at Thor. Loki had picked up his sword, and the two laughed together, their faces slightly obscured by the fur. As soon as Loki’s dagger had gotten between Thor and Thor’s ribs, they both pulled it out of their bodies, and Thor grinned. He reached out and took Loki’s hat off, and placed it on the floor and threw it at Loki. Loki caught it, but Thor cut his hat off and flung it behind hThe Enterprise dodged a laser blast and accelerated hard. The first few miles were uneventful, with no major damage. The next couple miles were more challenging, stretching nearly seven miles at a time. About halfway through the mile, however, something happened to the Enterprise that made it wobble a bit (this is often known as a change in direction) during each change. So to the right of the Starboard Front, the En
Missing some Tweet in this thread?
You can try to force a refresh.

Like this thread? Get email updates or save it to PDF!

Subscribe to Janelle Shane
Profile picture

Get real-time email alerts when new unrolls are available from this author!

This content may be removed anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

More from @JanelleCShane see all

Profile picture
Janelle Shane
@JanelleCShane
I used @runwayml to reconstruct this scene from The Last Jedi by having DeepLab recognize all the objects in each frame, and then having SPADE try to paint the original frame from the objects.
@runwayml More on the reconstruction process here: aiweirdness.com/post/186539822…
This is the first step, where DeepLab segments an image into objects. "Lightsaber" is not a category it has, so it improvises. luke skywalker holding a lightsaberluke is labeled person, background is mostly labeled wall, and lightsaber is labeled baseball bat
@runwayml Next, SPADE has to paint a picture just from DeepLab's labeled image. It sees person with a baseball bat, so it looks a lot like it tried to paint Luke in a baseball jersey. luke is labeled person, background is mostly labeled wall, and lightsaber is labeled baseball batluke has a blob of a face, a pixelated tshirt, and blue hair
Read 10 tweets
Profile picture
Janelle Shane
@JanelleCShane
So excited our paper is out! And written up by @carlzimmer! I personally calibrated the device in the picture, a programmable hologram that steers laser beams to zap neurons.
Our collaborators at Stanford eavesdropped on neuron activity while mice were seeing vertical stripe patterns. When they later displayed nothing but activated some of those same neurons by zapping them with laser beams, the mice acted as if they were still seeing stripes.
They don't know if the mice were seeing the original stripe pattern exactly. Maybe they were only seeing something that resembled stripes enough that the mice could figure out that they were supposed to do their pipe-licking thing, just like with real stripes.
Read 9 tweets
Profile picture
Janelle Shane
@JanelleCShane
7,500 pictures that image recognition algorithms tend to get wrong. My kind of dataset.
Interesting examples of image recognition algorithms being fooled by color, texture, and background cues.
They also seem to have learned that hummingbird feeder = hummingbird Classifiers may use erroneous background cues for prediction. Pictures of nails tend to<br /> co-occur with a wood background. Likewise, hummingbird feeders tend to appear in images with<br /> hummingbirds but are themselves not hummingbirds.
The image recognition AIs have also learned that bars = rocking chair, shadows+ground = sundial, and thin lines = harvestman spider.
Cruelest of all is the fact that there's no car or van option in ImageNet, but there IS limousine & sports car. Every ride is super fancy.
Read 6 tweets

Related threads

Profile picture
Ross Wilson
@rossalexwilson
So I’ve spent this evening touring BT Dial House in Manchester. It’s a huge telephone exchange, and brings telephony and Internet access to Manchester. Wanna see some photos and nerd out with me? A thread ... #internet #telephony #infrastructure
This is just one of the marine diesel generators that live in the lowest basement. The site has its own dual redundant HV feed, and operate their own HV/LV transformer kit. They take ~600 litres of oil, and are regularly tested.
To power those, they need a load of fuel. There’s a tank like this for each generator, and they hold enough red diesel for 8 hours. An additional huge set of tanks sit behind which receive fuel deliveries from a tanker connection outside the building.
Read 10 tweets
Profile picture
⭐️⭐️⭐️🇺🇸IN THE END WE WIN! =COVFEFE
@QsentMeToo
So many people don't know what Q is or even understand how to read it, so I will try to help you all. Q is NOT the news, not some crazy person who has a tin foil hat or conspiracy theorist, Q is TRUTH. It's FACTS presented to anonymous computer geeks who love to look through the
available government documents known as FOIA documents that are based in 100% FACTUAL data. Anons is a term used to describe the people searching and posting supporting DATA on the 'drops' of information known as 'CRUMBS' to those who 'BAKE' these crumbs into 'BREAD' or real time
headline information everyone can use to make up their OWN minds. Nobody who is a REAL anon will fault you or call you names for believing or not believing any of it because all true anons enjoy a debate on the facts and love to hear opinions. Sure, you may experience a shill or
Read 17 tweets
Profile picture
Darren Mooney
@Darren_Mooney
Still on a bit of a nostalgic #DoctorWho kick, in particular for the Moffat Era.

So now watching “Dark Water”, the first part of the two-part eighth season finale.
“It's a terrible thing. Just a terrible, terrible thing.”
“It wasn't terrible.”
“Clara?”
“It was boring.”
“Boring?”
“It was ordinary.”

Clara is a fascinating companion, in large part because she’s essentially riffing on the sense of entitlement that Rose felt.
“He deserved better. And so did you.”
“I don't deserve anything. Nobody deserves anything.”
“But I am owed better. I am owed.”

Clara is selfish and self-centred. But she’s aware of that, and the show is aware of that.

These traits are explored as developed. They make her human.
Read 222 tweets
Profile picture
Bart Dessaint
@bartdessaint
1/ #smarketplace i.e. the intelligent next generation of exchange platforms

New post I wrote in @TechCrunch: techcrunch.com/2018/10/23/sma…
2/ Marketplace businesses are intrinsically linked to the technologies that enable them. e.g. there would be no Uber without location services and ubiquitous smartphones
3/ As enabling technologies have evolved, marketplaces have grown to facilitate increasingly complex exchanges in new environments and industries. e.g. from Craigslist text listings to social network proofed + photo documented crafts Etsy.
Read 22 tweets
Profile picture
Alex J. Champandard
@alexjc
Neural approaches to style transfer struggle with certain types of art, e.g. crisp yet smooth brush-strokes 🖋️. It's likely a combination of factors, including using models pre-trained on natural images. 📷

In this thread I'll experiment to learn more! 👇 #neuralimagen #procjam
Ah. Here's what happens visually when the optimization algorithm (L-BFGS) "explodes" 💥. I have not tracked down the cause yet, likely extending #PyTorch to understand parameter ranges would help.

(Left: previous iteration looks OK, Right: pixels go far out of range → clamped)
Example outputs of the Picasso drawing with more bias towards upper layers of the convnet (4x), then equalized, then bias towards lower layers (4x).

Interesting, but nothing like the original. Some insights forming!
Read 31 tweets
Profile picture
dell cameron
@dellcam
👀👀👀
still updating this story as I go through the report, but some interesting interviews w/ senior FCC officials who disagreed w/ the "DDoS attack" narrative and said as much at the time.
Here's the former CIO of the U.S. (now an FCC advisor) saying he disagreed w/ the way the FCC described the purported "attack" publicly & that FCC's chief security officer also "saw the event as a resource capacity issue, and not a
malicious or DDoS issue" gizmodo.com/heres-the-inte…
Read 13 tweets

Trending hashtags

#Mobile #Perfidy #MBep7 #NotTrue #database #NeuralDoodle #DC #smartciities #GoodbyeDemocratsPatriots #Decentralisation #smarcities #WalkAway #PyTorch #nonenfait #USA #China #smarticities #startups #Falklands #Democrat #opendats #suicided #Dishonored2 #ai #Buildings
Follow Us on Twitter!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just three indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!