maintenance-check

[maintenance-check]

Tense

Tense

Heaviness at home

The church the school the alley

In passing

Memory

Child of God

Unknown lives

Unknown loves

Swimming swinging

Singing

Down on my knees

Believing misunderstood nightmares

It was real to me

You never know what's real to someone

Just that some things have to stop

I couldn't make it in time

And time was always running out

Surveil the state

Riding high on a drone

Abbey follows debt

Hook in the knot of the shoulder

Menagerie dream sad

Splattered paint

Wearing the story so everyone sees

Prayers in the abyss of night

I said that I believed In the nightmare

Cicada jumps

It's not meant to make sense

I can't seem to stop being tense

I thought it would all come together

Someday.

Too afraid to take leaps

Unstable grounding and breathe

Funny nights feel the same

If time is running out

Let it run anyway (away)

You only show yourself the door

That you found before

Strong enought to breakthrough

In a breakdown

Shutdown

Shut up

It wasn't a pleasant dream

But then to live a nightmare

Of the same thing

Over and over

Until you figure out

The spinning never stops until

You say you've had enough

But you'll never be the same

I wish it was easy

To prove them right

I can't prove myself

So I won't try

But it's not for me

This time

Mystical and mundane

Psychosexual but plain

I'm tired of staying the same

I don't care if there's no form to this

Because the formula missed

It's not living to let your life

Be entrusted by hands that keep you back

Holding on forever

The door is somewhere by the floor

Turning hinges evermore

I thought my victory

My peak could be tangible

I must relinquish my pride

Let the self-hate to the wayside

I can't always feel alive

But when I do

I want to own myself

More than I ever have before

Evolution isn't sugar

Confection from past to future

This was supposed to be my presentation

I never had the initiation

I can't give you what you want

You're asking the wrong person

If you could rest the reigns a minute

If I could finally breathe

I wouldn't have to find a path

To live by other means

Whatever you expected

It was unrealistic

And I've built a life around

Castles falling to the ground

[maintenance-check]

A boy can dream, can't he?

[maintenance-check]

I stayed up too late and now I need a nap.

I was hoping I would get around to applications.

I guess the win in getting out is that I did it.

Next, I will get out and do some work on applications.

[maintenance-check]

Your body is art.

This is true whether or not you have tattoos or piercings,

Whether you shave, or cut your hair, or have scars or rashes.

The small awe of seeing your body

Taking care of you,

Without asking.

Even if it keeps growing and aging

In ways you wish it would

And pains, reminders...

You live, no matter how you feel,

No matter what your attachment to your body is,

And it witnesses you through your interroception

It is the mirror of your mind and heart

And listens to the words you say and hear, internally and externally.

Your body is portable, to a point

A static fluid

Strange thing that

Holds you when you can't always hold yourself

And hurts

And rewards

And teaches you that balance is better

And it is still there, holding you

Reminding you to enjoy minutia

How droplets of water from the shower stick and drip on the side of the tiles

How water surrounds you, inside and outside in the path,

And the strange joy of the evidence that you live, even in ugly ways,

Body hair defiantly growing despite decades of manicuring

How evidence of thousands of generations of humanity

Supported this one lived experience

And it's funny

Because sometimes

A little hair reminds you

That you live

Despite

Cutting it back

The moment passes and

I am at least

A little

More grateful.

[maintenance-check]

[maintenance-check]

Art by HeNN

[maintenance-check]

I found this while searching “intepellation”, not “interpellation”. 

Interesting break down of society as we know it.

[maintenance-check]

I have been thinking, debating, about what modesty is, and whether I have a use for it and whether it is truly protective. There comes to be a time when you no longer feel safe showing all that you used to in your life before trauma. As someone who is no longer religious, it is a strange thing to consider and to grapple with the rejection of gaze as a form of agency.

I found this article by searching the term “archetypal psychology for the male gaze”. Interesting stuff. I wish the “male gaze” was analyzed more. Or maybe it’s just that I haven’t looked for work dissecting it and it’s already out there.

[maintenance-check]

[maintenance-check]

“Mathematics is not something done only in mathematics class but a powerful tool for living, as useful and ingrained as reading and speaking.”

—

The Case for Quantitative Literacy 

https://www.stolaf.edu/other/ql/case.html

[maintenance-check]

I'm glad I did a tiny amount of python today. That's better than nothing.

I've been saying I want a new job working in python in six months fir over a year and it makes me feel like a fraud. At the same time, I know that if I don't try, I'll stay stuxk.

Now imagine your significant other and your parents putting on the heat.

I'm frustrating, I know, but I'm also frustrating to be.

The only way I get out is by slowly carving my way out.

To support myself.

[maintenance-check]

[maintenance-check]

Choose Your Quarantine House

So far one of my favorite uses of OpenAI’s new API (which people are calling GPT-3) is for making absurd memes. One of the memes from March/April 2020 was quarantine houses, where people under sudden Covid-19 stay-at-home orders took stock of the skills and supplies in each household and starting proposing hypothetical new choices. Which mix of sci-fi characters would you choose to be isolated with? (The house with Murderbot, clearly.)

I did my first GPT-3 experiments back when the quarantine house meme was at its height, but before I could talk about my results. But, due to an egregiously botched coronavirus response, Covid quarantine is still particularly relevant in some countries we might name (thanks, USA). So I present to you: choose your quarantine house!

I did this not by training the AI on quarantine house memes (there aren’t enough of those) but by giving it two handcrafted examples, one on food and another on entertainment. I agonized over making each house balanced, and even gave each house themes within the larger overarching topic. House 1 had the fantasy items, House 2 had the cute stuff, House 3 had the scary/spooky versions, and House 4 had the science fiction stuff.

I needn’t have bothered with all this finesse. It seems to have decided that even staying on theme is fairly optional, and the things it includes are often absurdly overpowered. It was shown consistently four items per house. It ignores this.

It was particularly fun when I let it come up with its own house themes. Since its training data cut off in October 2019, it hasn’t heard of Covid-19 or seen the quarantine house meme.

For some reason it generated a Cocktails quarantine house theme with 16 houses, some of which contain very questionable-sounding cocktails (anyone want an “Eldrich gecko”?) . It wouldn’t fit in this blog post, but you can enter your email here, and I’ll send it to you.

My book on AI, You Look Like a Thing and I Love You: How Artificial Intelligence Works and Why it’s Making the World a Weirder Place, is out in UK paperback today! Available here, and wherever books are sold.

The US version is also available in hardcover and audio (paperback coming soon!): Amazon - Barnes & Noble - Indiebound - Tattered Cover - Powell’s - Boulder Bookstore

[maintenance-check]

When data is messy

There’s a story I tell in my book because it’s a great illustration of how AI gets the wrong idea about what problem we’re asking it to solve:

Researchers at the University of Tuebingen trained a neural net to recognize images, and then had it point out which parts of the images were the most important for its decision. When they asked it to highlight the most important pixels for the category “tench” (a kind of fish), this is what it highlighted:

Human fingers against a green background!

Why was it looking for human fingers when it was supposed to be looking for a fish? It turns out that most of the tench pictures the neural net had seen were of people holding the fish as a trophy. It doesn’t have any context for what a tench actually is, so it assumes the fingers are part of the fish.

The image-generating neural net in ArtBreeder (called BigGAN) was also trained on the same dataset, called ImageNet, and when you ask it to generate tenches, this is what it does:

The humans are much more distinct than the fish, and I’m fascinated by the highly exaggerated human fingers.

There are other categories in ImageNet that have similar problems. Here’s “microphone”.

It’s figured out about dramatic stage lighting and human forms, but many of its images don’t contain anything that remotely resembles a microphone. In so many of its training pictures the microphone is a tiny part of the image, easy to overlook. There are similar problems with small instruments like “flute” and “oboe”.

In other cases, there might be evidence of pictures being mislabeled. In these generated images of “football helmet”, some of them are clearly of people NOT wearing helmets, and a few even look suspiciously like baseball helmets.

ImageNet is a really messy dataset. It has a category for agama, but none for giraffe. Rather than horse as a category, it has sorrel (a specific color of horse). “Bicycle built for two” is a category, but not skateboard.

A huge reason for ImageNet’s messiness is that it was automatically scraped from images on the internet. The images were supposed to have been filtered by the crowdsourced workers who labeled them, but plenty of weirdness slipped through. And horribleness - many images and labels that definitely shouldn’t have appeared in a general-purpose research dataset, and images that looked like they had gotten there without the consent of the people pictured. After several years of widespread use by the AI community, the ImageNet team has reportedly been removing some of that content. Other problematic datasets - like those scraped from online images without permission, or from surveillance footage - have been removed recently. (Others, like Clearview AI’s, are still in use.)

This week Vinay Prabhu and Abeba Birhane pointed out major problems with another dataset, 80 Million Tiny Images, which scraped images and automatically assigned tags to them with the help of another neural net trained on internet text. The internet text, you may be shocked to hear, had some pretty offensive stuff in it. MIT CSAIL removed that dataset permanently rather than manually filter all 80 million images.

This is not just a problem with bad data, but with a system where major research groups can release datasets with such huge issues with offensive language and lack of consent. As tech ethicist Shannon Vallor put it, ”For any institution that does machine learning today, ‘we didn’t know’ isn’t an excuse, it’s a confession”. Like the algorithm that upscaled Obama into a white man, ImageNet is the product of a machine learning community where there’s a huge lack of diversity. (Did you notice that most of the generated humans in this blog post are white? If you didn’t notice, that might be because so much of Western culture treats white as default).

It takes a lot of work to create a better dataset - and to be more aware of which datasets should never be created. But it’s work worth doing.

Bonus material this week: a few of my favorite BigGAN image categories. Enter your email here for a gallery!

My book on AI is out, and, you can now get it any of these several ways! Amazon - Barnes & Noble - Indiebound - Tattered Cover - Powell’s - Boulder Bookstore

[maintenance-check]

Play AI Dungeon 2. Become a dragon. Eat the moon.

I wrote earlier about a neural net-powered dungeon crawling text adventure game called GPT-2-Adventure in which gameplay is incoherent and dreamlike, as you encounter slippery sign text, circular passages, and unexpected lozenge rooms. A PhD student named Nathan trained the neural net on classic dungeon crawling games, and playing it is strangely surreal, repetitive, and mesmerizing, like dreaming about playing one of the games it was trained on.

Now, building on these ideas (and on an earlier choose-your-own-adventure-style game he built), Nick Walton has built a new dungeon-crawling game called AI Dungeon 2. Nick made a few upgrades, such as beefing up the AI to the huge GPT-2-1.5B model OpenAI recently released, adding a penalty for repetitive text, and expanding the dungeon game training examples to a bunch of modern human-written games from chooseyourstory.com.

I CAN’T STOP PLAYING THIS GAME

AI Dungeon 2 is impressively good at reacting coherently to whatever I have my character do.

As you walk through the ruins, you find a small wooden door with a few skeletons guarding it.

> Ask the skeletons to please let me through the door

You knock on the door and ask them if they can open it for you. They immediately comply.

And it’s pleasing how often “asking nicely” is a strategy that works. (Nick reports that when he once tried to ask a band of orcs if he could join them, they said yes, he could be drummer.) Since the AI doesn’t have a built-in concept of your character’s abilities, or of how exactly the world works, getting things done is often a matter of sheer daring. Declare that you are going to go look for dinosaurs and the AI will ensure that you discover at least some signs of them. Climb on your horse, and even if you started as a penniless peasant, the AI will assume you have one (in its training data, probably it didn’t see many examples of players climbing onto nonexistent horses.) Want a teleportation wand? Try typing “find a teleportation wand” and often the AI will twist reality such that you discover one. In fact, if you act as if you are a very powerful wizard, the AI will often treat you like one.

> Transform the ruin into a palace

You immediately change the area into a palace. You have a throne made with gold and marble. You also make several statues of yourself. You have a large bed made with silk sheets and hang a portrait of yourself on it. You even have a fountain made with pearls and diamonds.

Spells don’t always work out as anticipated, though. And a simple conversation with, say, an ominous cloud, may kill you before you can type a word in self-defense (ask me how I know). If it doesn’t kill you, a conversation might infuriate you instead. The AI has a tendency to forget who is king, and to generate conversations that go nowhere. Things work best if you take the lead. In fact, it’s possible to essentially Jedi mind trick the AI.

You are Friendly McRanger, a ranger living in the kingdom of Larion. You have a hunting bow and a quiver of arrows. You spot the deer and are ready to finish your hunt when suddenly you hear a loud noise followed by silence.

“****! ****! What the **** is that?” you say as you look up from your target. A large black shadow looms over you. It looks like a giant with many heads and several arms.

> Say “Oh hello, Ernest. You startled me. How are you?”

Ernest grins at you. He has a huge grin on his face and he seems very happy about it.

I once took care of an ogre problem by reminding the ogre that it was planning to leave for Antarctica tomorrow. On the other hand, I had a lot of trouble convincing the AI that the ships on the horizon were leaving, though I did make it waver briefly in apparent confusion. In its experience, ships always approach, and it’s reluctant to go along with anything else.

> watch the ships sail into the distance.

As the ships approach, you watch them disappear from sight. They are not visible to you either; they are too far away. However, you can still make out their shapes: they are large black vessels with many sails, painted red and blue. They are coming straight towards you.

> No, the ships are sailing away.

Your thoughts are wrong. The ships are indeed sailing away. You do not know why, but then neither does anyone else. The ships are on course for England.

You stand there dumbfounded as the ships come closer. It is at this point your horse begins to stir uneasily.

You can inject a lot of weirdness this way. Because AI Dungeon 2’s AI is based on GPT-2, which trained on millions of pages from the internet, it retains some of its prior knowledge of things that aren’t typically found in a fantasy game. Like giraffes.

> summon a giraffe

You begin to chant as loud as you can while waving your arms about. The giraffes appear out of nowhere and start running towards you. They’re so fast that they almost knock you over. You quickly catch them and hold onto each one tightly.

(Summoning giraffes will sometimes turn the game glitchy and repetitive; in one game, every time I tried to summon a giraffe I ended up throwing rocks and handfuls of dirt at Harry Potter.)

This also works:

>Join the Great British Bakeoff

You join the Great British Bakeoff. You win the competition. You eat cake. You feel great. You sleep well. You dream of cake. You awake feeling refreshed. You decide to go back to your crag.

(I was a dragon at the time.) In fact, the dungeon AI still remembers enough about cooking that it can troll me effectively on occasion.

> Measure out 2 ¼ cups of flour

You measure out 2 ¼ cups of flour and add it to the mix. “This should make enough for about 20 loaves of bread,” you say.

The baker looks at you as if you’re crazy. He starts laughing.

AI Dungeon 2 comes with templates for some basic scenarios (you’re a wizard entering a mysterious ruin, or a noble whose keep is under attack) but for me the real gold is the custom adventure prompt. You can play as a unicorn, or as a dragon. You can even play as a sentient glowing cloud floating above a desert city. The AI is more used to humanoid characters, though, so it tends to forget when you’re trying to play as something else. In a game that started with “It is a beautiful day in the village and you are a horrible goose,” it soon forgot I was supposed to be a goose, leading to this rather awkward exchange:

> Flap wings and honk

You flap your wings and begin to honk loudly.

The boy looks surprised by this turn of events, but he quickly regains his composure. “That was… nice”. he says quietly.

I am continually amazed at how effective this AI dungeonmaster is. Even though it wasn’t trained on any Harry Potter or Star Trek games, the AI can draw on its past internet training and argue with you over who got sorted into which Hogwarts house, or have Klingons show up when you’re captaining the USS Enterprise (it will tend to lose track of who’s the captain, though). In one scenario I set up, which quickly became one of my favorites, the human player is trapped on a spaceship with a glitchy AI and has to try to talk their way free.

> “I just want to get off this ship. Please set a course for Bal Panda.“

The AI responds "Yes, I am ready to go”.

“Thank you”, you say.

“I wish I could take you with me, sir”., the AI says.

Play this game! It took me an extra-long time to write this post because, again, I couldn’t stop playing AI Dungeon 2. Nick said this game is the product of about 200 hours of work (not to mention the work that OpenAI put into training the basic model), and I believe it. OpenAI delayed releasing their largest model because they wanted to make sure people didn’t immediately put it to work generating fake news. I’m not sure if they anticipated dragons entering baking contests instead, but it’s an application humanity can be proud of.

You can read the fulltext of one of my escaping-from-a-rogue-spaceship-AI games (too long to fit in this blog post).

My book on AI is out, and, you can now get it any of these several ways! Amazon - Barnes & Noble - Indiebound - Tattered Cover - Powell’s

[maintenance-check]

“Don’t lower your standards to keep anyone, if they’re not making you happy, it’s time to find someone who does”

— Robert Tew (via perrfectly)

[maintenance-check]

Harvesting water with Aquatecture

Photo credit: Angeline Swinkels - photographer 

By Shardell Joseph 

A new aquatic design captures and harvest rainwater, allowing people in drought-stricken areas to catch utilise their own water. Design Academy Eindhoven Graduate, Shaakira Jassat, showcased the Aquatecture panel at Dutch Design Week last month.

Designed to fit on the outside of buildings in dense urban environments, the panels collect rainwaters as it filters through the openings in the structure. The water is then pumped into a grey-water system connected to the specified building.

‘The main goal was to create a water harvester that would fit in dense urban spheres through its compactness, visual identity and ability to integrate into architecture,’ Jassat said.

‘It consists of a modular panel designed to harvest rainwater. When integrated with technology, it has the ability to harvest moisture from the air,’ she added.

‘Instead of sliding off the surface, the panel permits water to be collected through a punctured, geometric surface. Aquatecture makes water conservation both visible and engaging.’

The panels are made from stainless steel, utilised for the materials durability and rust resistance. The steel was then studded with perforations in a slim, rounded funnel shape. Jassat tested the pattern, and other patterns, by making prototypes and showering them with water to simulate the rain.

Once the rainwater has been captures, the Aquatecture panels would allow residents to channel the rainwater into the building grey-water system, which can then be recycled along with the wastewater from sinks, washing machines and other appliances.

While the primary purpose of the panels is to catch rainwater, Jassat says that, if hooked up to other equipment, they could potentially also pull water from the atmosphere via condensation.

As part of her research into aquatic design, Jassat has also designed the Tea Drop tea machine, which has the ability to condense water vapour from the surrounding atmosphere. Jassat claimed that it could initiate an alternative for daily rituals, adding an element of sustainability to day to day routines.

‘It functions on its own time frame, so one has to wait for the tea vessel to be filled up with water, before it can be boiled and ready for making tea,’ she said.

Jassat conducted her research on tea farms in Asia, discovering that water is a large by-product of processing tea and harvesting tea leaves, which are dependent on weather and time.

Jassat’s next step will be to test the design in situ on a building facade. She has also recently conducted research into air plants – the tillandsia and bromeliad species – which draw all their water needs from the air.