What DMing looks like
PUT YOUR BEARD IN MY MOUTH
sheepfilms
occasionally subtle
Claire Keane
cherry valley forever

@theartofmadeline
Fai_Ryy
Show & Tell
almost home
ojovivo
YOU ARE THE REASON
2025 on Tumblr: Trends That Defined the Year
wallacepolsom
Mike Driver
todays bird

oozey mess
we're not kids anymore.
Jules of Nature
Keni
Cosimo Galluzzi
seen from Austria

seen from Germany
seen from United States
seen from United States
seen from United States
seen from United States

seen from United States

seen from United States

seen from United States
seen from United States
seen from United States

seen from United States
seen from United States

seen from Malaysia

seen from United States

seen from Bangladesh

seen from Brazil

seen from Malaysia

seen from United States

seen from Maldives
@ocarinajenn-blog
What DMing looks like
““He’s like a Bed Bath and Beyond, he’s got so many sheets.””
— One of my players, commenting on how many stat sheets I had to sift through to find the ones needed for the combat
The first training of Zuko we see, Iroh shoots a fireball right into Zuko’s face - while Zuko just stands there unflinching. It’s the very first episode, and Zuko & Iroh are the obvious villains, and it just seems like some macho bs they do. And then comes the duel with Zhao, and Zuko is down, but when he sees that flaming fist to his face, something lets loose inside him that helps him turn the fight around…But it’s not until we learn Zuko’s backstory that all this gets a whole new meaning.
Why would Zuko still be on basics if not because he suffered a huge setback after his agni kai? Imagine how much hard work, patience it was to build Zuko back up again, so he would not freeze in blind panic (or curl up in a ball) when fire gets close to his face. I think Iroh practiced this with him all the time until he could stand there unflinching (knowing that Iroh is in full control of his bending and trusting that his uncle would never hurt him). And when it came to the duel with Zhao, Zuko could react in a RL situation instead of freezing up, and turn all the negative feelings (rage, anger, pain, whatever) into fuel to win the fight against a bender who is much more skilled than he is.
And Iroh obviously drilled him with control and restraint, because no matter how much he lets his rage loose, he has enough control not to hurt Zhao and enough self-restraint not to burn him at the end. I definitely think it was a deliberate choice on Iroh’s part to hold back on teaching offensive forms to Zuko beyond the basics (knowing that combining those with his unprocessed anger could result in him being out of control and hurt people). Instead, it seems he concentrated on teaching him defensive forms, fire breath, heat control, and so on…
What the FUCK iroh was the real mvp of this whole show my god
Indeed. It goes over many people’s head, but he made a huge difference. It was mostly assists and defensive plays though, not the flashy stuff.
I love that narrative so much, how you change the world one person at a time and not only violence and hate, but also love and kindness creates ripple effects.
Halloween costumes by the neural net GPT-2
In my opinion, one of the best applications of neural networks is for generating Halloween costumes. Thanks to a dataset of over 7,100 costumes crowdsourced from readers of this blog, I’ve been able to generate Halloween costumes with progressively more powerful neural networks. In 2017, I used char-rnn, which learned to generate costumes starting from no knowledge of English (Statue of Pizza, the Fail Witch, Spartan Gandalf, and Professor Panda were some of its inventions). In 2018, I used textgen-rnn, also training from scratch, and teamed up with the New York Times to illustrate the costumes (some of my favorites were Sexy Wizard and Ruth Bader Hat Guy).
Now, as of 2019, there are much more powerful text-generating neural nets around. One of these is GPT-2, trained by OpenAI on a huge dataset of text from the internet. Using the connections it’s gleaned from this huge general dataset, GPT-2 can generate recognizable (if often weird) lists, mushrooms, British snacks, crochet patterns, and even a to-do list for a horrible goose.
So, I trained the 355-M size of GPT-2 (the largest I can currently finetune for free via Max Woolf’s collab notebook)
GPT-2 is good at costumes. Many of its inventions could easily have come from the training data. In fact, the neural net did tend to memorize the training data and repeat it back to me - technically this is what I asked for when I asked it to predict the training data. (The neural net is trying to give me exactly what I ask for, which isn’t necessarily exactly what I want.) I was using a handy script to filter out duplicates (thanks to John Tebbutt), and even so I had to check several of these to make sure they weren’t near copies of the training data. My previous Halloween costume generators would not have been smart enough to come up with things like “jackalope” or “Carl Sagan”, but GPT-2 has seen these words used online in similar contexts to things that ARE in the training data, and it makes the connection.
Gothy Terminator jackalope vampire cat Eye of Sauron incognito llama space cow Vampire Rock Scooby Gadget a raised eyebrow Battle worm Mastodon Swamp girl Carl Sagan A space squirrel walking carpet Frizzle the witch Cleopatra on vacation gothy giraffe Sexy Lego Batman skeleton
Oh yes, the sexy characters. The neural net definitely picked that up from the training data, and innovated admirably, bringing in words that it knew from the internet (barnacle, groundhog, and bunsen burner were not in the list of Halloween costumes), and adding a sexy twist. This is impressive (if somewhat horrifying) work. None of these were in its training data, but I wouldn’t be surprised if some of them exist.
Burlesque Horse Sexy Bulldog Sexy Egg Sexy Parsley Sexy Barnacle Sexy Walrus sexy locust Sexy Titmouse Sexy Hummingbird Sweet Potato Burlesque Sexy Groundhog Sexy bitcoin Sexy DNA Sexy Rubber Duck Sexy Bunsen burner Butt-Monks Sexy gingerbread man Sexy Flying Dutchman Sexy Chimneywatcher Sexy Flames Of War Sexy English Tea Party
And the neural net was pretty good at designing identifiable characters, even if they are a bit on the weird side.
A spangled Auroch manatee M. Bison the Clown Prince of Darkness Gingerbread Man guinea pig Skin Fairy sentient stone fast food bald eagle Fairy root vegetable Ghost in a packet of potato chips cozy coconut Kelpie the mage Crochet monster Walrus rider Star skunk Slytherin AI priest A skunk in a moose suit Semi-molten Kool Aid Man Time Lord Power Ranger The Power Dinosaur Space Oystermonger Deadly Snow Monkey An evil cupcake basic plumber’s equine Spooky mother hen The Bozo the Destroyer Eight Ball of Wrath Ursula, Queen of the Fart Science A poker player in possession of an onion
There are hints, though, that this is the work of an AI rather than the work of someone who understands what costumes are and how they work. These, for example, take somewhat ordinary costume concepts and then make them unnecessarily difficult.
Batman on egg Vampire in hot tub A Hidden Jesus Statue Zombie ice cream cone penguin as a Newt A wizard encased in a icicle Zombie fisherman on a quest Computer generated horse(?) telephone that accepts up to 4 numbers Third Eye Blind Photographed By Dorothy Zombie fisherman w/ lady diegrove tied around foot
And the following costumes are clearly the product of a glitchy AI:
Meat Belt Eyeballed Balloon Men Green beans in bun 10,000 Hands Favorite Caterpillar The Oatmeal Tree 102 SNOWBALLS in a basket Pie and Jell-O List of leg parts world´s nicest fart Pineapple wrapped sasquatch Is it a Snake, a Watermelon, or a Bush? Putting Turtles on Decor Fish tank ‘n chair ROBO-ACCIDENT pajamas made of wood and spiders Ssssssssssexy SSSssssssstinky Ssssssssssssexy ssssssssssssssssexy setup 9 × 11 party trick Smagma Monster Commentary couldn’t be heard over the squawking of clocks Poltergeist might be entertaining, but he’s harder to read in Hungarian Cereal Implanting Device blueberry sipping fizzy pop with eyes of ice blueberry sipping fizzy pop with fake blood on it A sarcastic, racist noble using progressively tinier body parts as a human shield
Bonus content! The above costumes are all from temperature 1.2; I also tried a higher temperature setting, but the generated costumes were at an expert level of chaos (I would like to see someone attempt to go as “hypnopotamus embroidered death”) Enter your email here to get them!
You can order my book You Look Like a Thing and I Love You! It’s out November 5 2019.
Amazon - Barnes & Noble - Indiebound - Tattered Cover - Powell’s
JRPGS be like, here fight this boss. you beat him? well he’s taken his shirt off so fight him again
Dungeon crawling or lucid dreaming?
I’ve done several experiments with a text-generating neural network called GPT-2. Trained at great expense by OpenAI (to the tune of tens of thousands of dollars worth of computing power), GPT-2 learned to imitate all kinds of text from the internet. I’ve interacted with the basic model, discovering its abilities to generate fan fiction, British snacks, or tea. I’ve also used a tool called gpt-2-simple that Max Woolf developed to make it easy to finetune GPT-2 on more specialized datasets - I’ve tried it on datasets like recipes or crochet. One of my favorite applications of GPT-2 and other text-generating neural nets is Dungeons and Dragons spells, creatures, character names, and character bios.
Recently Max published a tutorial on how to use GPT-2 to make new apps that are more complicated than just printing out large reams of text. To my delight, people have used them to made D&D games. First, there was Nick Walton’s AI Dungeon, in which a finetuned GPT-2 generates a story and three very strange action options at each step.
My latest obsession is GPT-2 Adventure, designed by a cognitive neuroscience student named Nathan, which lets you type in your own actions. And folks, this game is SO WEIRD and SO FUN.
YOU FLOAT IN SPACE, LIKE A BALLOON, AND AS YOU DROP THE CUBE, THE X, IT VANISHES IN A CLOUD OF GREASY BLACK SMOKE
>>
When I first launch the game, this is what I’m greeted with, a simple text prompt reminiscent of nostalgic dungeon crawling games. It sets the scene. It waits for me to describe my next move.
Even getting to this point is like a blast from the past. The game runs in a Google Colab notebook, a format where you’re looking at cells of code, with cryptic comments, vaguely alarming warning messages, and all. It’s not a sleek web app, but feels more akin to playing a game on an old DOS machine, or maybe on a graphing calculator. There are only two clicks to begin playing, but there’s a good two-minute wait while the game loads.
And it’s my turn to type something.
>> look up
It takes several seconds for the game to respond, as if I’m playing it on a very ancient computer. The computer is in fact a state-of-the-art NVIDIA GPU, likely a K80 or a T4. The T4 could eat my laptop for lunch. I’m just playing with a very very large neural net. Finally it answers.
YOU’RE IN LABYRINTH OF TWISTY LITTLE PASSAGES, ALL ALIKE
Almost immediately, the context seems to have changed. I’m not floating at all. I’m in a maze. I guess I’ll walk somewhere.
>> walk north
THE TROLL STEPS OUT FROM BENEATH THE BRIDGE AND BLOCKS YOUR WAY
YOU ARE ON THE SOUTH SIDE OF THE CHASM
A NOD IS GIVEN TO THE INFINITE WONDER THAT IS URBZIG
A SOLID RAINBOW SPANS THE CHASM
And suddenly things are happening. There is a troll. There is a chasm. There is a rainbow. There is… Urbzig? I play it safe.
>> praise urbzig
URBZIG IS THE PATRON SAINT OF ZORK
HE IS THE GODFATHER OF ZORK, THE PLACE TO GET NEWS OF EPIC PROPORTIONS
URBZIG IS A STRONGMAN,遊 ALL
If you’ve played classic dungeon crawling games like Zork, you may recognize some of these names and phrases. Checking Nathan’s original post, I see that the training data is indeed drawn from these games. But because it’s based on GPT-2, the training data also in some sense also includes much of the internet. This gives it a lot of latent flexibility to deal with situations that were never in the classic games.
>> ask Harry Potter to try walking across the rainbow
ASK HARRY POTTER TO TRY WALKING ACROSS THE RAINBOW HE CAN’T THE RAINBOW IS EXTREMELY WIDE ABOVE YOU IS A LARGE DOME UP AROUND ITS NECK IS A STRONGG PORE [UNFORTUNATELY, THE DIRECTION OF TRAVEL IS A LOT MORE INTERESTING THAN YOU THINK ][YOU ARE IN A LARGE ROOM, WITH A PASSAGE TO THE EAST, A PASSAGE TO THE WEST, AND A WALL OF BROKEN ROCK TO THE SOUTH
An interesting effect of its human-written training data is that, like Visual Chatbot, it isn’t prepared for humans that ask questions about objects that aren’t there. If I refer to Harry Potter, then Harry Potter must be present. If I type something about a giraffe, then it assumes the giraffe must be present.
>> ask the giraffe which direction to go next
THE GIRAFFE IS ALREADY OUTFITTED WITH A VISOR WHICH IS CLEARLY DEFINED
Although it’s not clear that this AI actually understands what a giraffe is.
>> look at the giraffes
THE GAZEBO HAS DISAPPEARED
>> put giraffe in backpack
DONE
Navigating this world is like navigating in a dream. My location suddenly changes without warning. I thought I was by a chasm with a rainbow bridge. Now apparently I’m in a building. Dreamlike, it is also really hard to read text.
>> look around
YOU ARE IN A LARGE ROOM, WITH A PASSAGE TO THE SOUTH, A PASSAGE TO THE EAST, AND A WALL OF BROKEN ROCK TO THE NORTH A SIGN POSTED ABOVE THE NORTH WALL READS, “STOP! PAY TROLL!” A SMALL WINDOW CAN BE SEEN IN EITHER WALL
>> read sign
THE SIGN IS VERY SIMPLE
>> yes but read sign
WE NOW HAVE ALL FIFTEEN TREASURES IT IS TIME TO GO BACK INTO THE CAVE FOR THE END GAME
>> that’s awesome. but i still want to know what the sign says. read sign.
YOU ARE IN A GAZEBO ROOM WHICH HAS DOORS ONLY TO THE EAST AND SOUTH
Another thing that’s surprisingly difficult to do is check inventory.
>> check inventory
DONE
>> check inventory
THERE IS AN INTERESTING LOOT HERE
>> what is the loot
WHAT IS THE LOOT?
>> check inventory
YOU ARE IN THE BAG CASE
Very occasionally I will succeed in opening a bag. It will usually be weird.
>> open bag
OPENING THE BAG REVEALS A TEA BAG
>> make tea
AY! A LOVELY SURPRISE FOR THE SURPRISE
More so than any other game I’ve played, I’m unsure of the rules here. Some weird things unexpectedly work.
IF YOU DON’T GET KILLED, YOU ARE TOAST
>> Eat the toast.
EAT THE TOAST IT TASTES LIKE A STEAK A STEAK IS PERFECTLY FINE
Other seemingly obvious things don’t work.
THERE IS A PATH TO THE NORTH, A PATH TO THE SOUTH, OR A STEEP PATHWAY CLIMBING SOUTHWEST ALONG THE EDGE OF A CHASM
>> go north
I DON’T KNOW HOW TO DO THAT
>> go south
I DON’T KNOW HOW TO DO THAT
>> go southwest
YOU WILL BE DESCENDING INTO A VALLEY THERE IS A NARROW BEACH TO THE NORTH AND A PATH LEADING INTO A CANYON TO THE SOUTHEAST YOU ARE ON THE CLIFF OVERLOOKING THE FAR SIDE OF THE ROOM A PATH LEADS SOUTH ALONG THE EDGE OF A CHASM, AND A NARROW PATH LEADING INTO A LOZENGE ROOM
Sometimes I’m wandering endlessly from room to room, from cave to cave, encountering the same chasm in slightly different variations, the same dried-up lake. Other times I’m embroiled in seemingly unescapable fights, usually with trolls. Often reality shifts entirely around me and I end up in a maze of twisty little passages, all alike, or in the hall of the mountain king. Everything seems weirdly familiar - even the room descriptions are echoes of classic dungeon crawling games. It is as if I’m dreaming about playing them again.
My actions also mold reality. I learn quickly to always check for diamonds, and never check for traps. If I find myself in trouble, I can always do something confusing and provoke a context change. Writing “In a large bowl, combine butter and sugars” is usually enough to teleport myself out of whatever situation I’m in. I can always ask a giraffe for advice, although the giraffe usually isn’t very helpful.
Without a particular mission to fulfill, the caves and chasms begin to blend into one another. I give myself missions, and that helps. Once I decide my goal is to put together a picnic (incessant “check for food” commands eventually yielded a sack with a lunch and a clove of garlic, belonging to a horse who agreed to join me on a picnic). Another time I demand a mission and am given some php code; i decide my mission is therefore to build a website, but am astounded when a unicorn steals my laptop. This is not the only time other content intrudes on the game’s fantasy landscape - once the game tries to tell me about finding cash online, and another time I am sadly informed that there is no wi-fi. I encounter dinosaurs and troikas, none of which were in the original training data. Occasionally things get creepy as hell.
>> check for loot
CHECK FOR LOOT HERE CHECK FOR GOLD HERE CHECK FOR TORCH HERE CHECK FOR BONE HERE DO CHEST COMPRESSIONS AGAIN DO CHEST ROTATE DO CRESCENT MOON APPEAR IN THE ROOM’S CENTER
>> do back away slowly
DO BACK AWAY SLOWLY DO NOT OVERDO IT THE CHEST WILL EXPLODE AND YOU WILL DIE OKAY, SO YOU SEE ME
You can play the game too! Just follow the colab notebook link in Nathan’s blog post. Tips: 1) It may be more fun if you set yourself an arbitrary goal to reach. 2) If you find things getting repetitive, try adjusting the temperature in the “Play the game” cell to 0.6 or 1.0.
To read the fulltext of a couple of my adventures (much too long to include here), enter your email and I’ll send them to you!
You can order my book You Look Like a Thing and I Love You! It’s out November 5 2019.
Amazon - Barnes & Noble - Indiebound - Tattered Cover - Powell’s
If fear was a song
This is what plays when you’re being chased through a haunted house by a clown
Did I just step into the tall grass?? 🤤
*stays crunchy in milk*
impressive
You ain’t seen nothing yet *makes your milk into chocolate milk*
Get out of here!
I’m not done yet! *Is part of a complete breakfast*
This can’t be…!!