
Subscribe to The Cheshire Cat to receive new posts directly to your inbox.
Over 100 subscribers
It was mid-December when the idea first came to me: Using agents for writing. But before you hit cmd+w to close the tap, bear with me. It's not another copy-writing productivity hack agent.
My vision is grand (as all visions are you mumble): An experience where LLM and people co-create a story, where everyone is equal and it feels like a meshing of minds. And where your creativity gets rewarded. Where writers don't just survive but thrive and where quality is king and speed is thrown out the window.
"AI is psychological asbestos", was the first comment I received after presenting my idea at a Kernel's open hour. I was taken a back, shocked at the person's ignorance of technological opportunities. But I needed to hear it to reconsider my position. It forced me to evaluate the (mis) alignment between my vision and my bot.
I was in week 2 of kernel, a month after my second hackathon submission (Worldbuilding bot in Telegram), three months after kicking off this project with a simple farcaster bot. And instead of taking a big leap forward, I was thrown back to square one. What the telegram bot did, doesn't matter anymore. I killed it. You always kill your darlings. I was immensely proud to pull it off, vibe coding and swearing at node and git while ignoring my Tinnitus ringing.
Before receiving this feedback, my plan for Kernel was simple: Just rebuild the Telegram bot for Farcaster. Then others will use it. I will get feedback, build in public and so on. But the world doesn't need more bland AI crap to flood our timeline. I know that. We don't need a LLM to write stories for us, we need ways to write stories together. For no other purpose than it is fun to make up stuff.
As a kid you dressed up, you played with Lego, dolls or cars. You shifted into character and imaged new worlds with your friends. And then you went to school. You were told to grow up, to behave, to not make fart jokes and to accept that 2+2 = 4 (it does, but only because of the social consensus that we use a 10-point number system).
You know the game where you say one sentence and the next person has to add to it? My Worldbuilding bot on Farcaster turned out to be similar to that game. Unintentionally. Someone mentioned I'm sorry did you say street magic ? a table top role playing game. I liked how it starts with players agreeing on three adjectives to describe the city they are building. It creates a playground biased by its players. By copying this design element, I made one crucial change to the flow: The story has to start with the human, not the LLM
In my Worldbuilding game, the foundation of the world is created through the LLM analyzing the user's posts on Farcaster and suggesting three adjectives. This isn't a random way of coming up with adjectives. There is a tremendous amount of a human's unconscious in the text. I'm not talking about the self-censoring we are doing. This is conscious. But without planning to do it, the words you select include signs of your gender, psychological traits, or even sexual orientation. Fascinating research I read back in 2015, when text research had a moment thanks to the rise of machine learning.
Currently, you can talk to Worldbuilding. It has two levels: entry and worldbuilding. To enter the Worldbuilding level you need to say the magic words. Once in it, you'll be given three adjectives (give it a second). And then it's up to you to start creating your world. You can have as many back and forth with the bot as you want. Multiplayer mode not yet available.
I had two early successes with the bot. First borrowlucid's statement that's fun and the person Susan (a person I created while writing a story to make debugging the bot more enjoyable) being picked up by jianbing.eth for 1st Farcaster-based comic /trending casts. I also got kid#2 to try it out. She was less impressed, as "the fun of writing poems is to write them yourself".
Another unintended side effect of the Worldbuilding LLM bot on Farcaster is that I wrote a story about humankind candles. Enjoy.
Title: Fragmented reality
Once there but suddenly gone, hope disappeared the second the TV shut off. She should not have smashed the lavender candle into the big new flash wall tv, Laura thought, but oh it felt so good!
"Maybe the TV could be glued back together?" Laura spoke to herself and picked up a piece. It was soft, she could bend it. Mesmerized she didn't realize how her arm was changing.
When Sarah looked at her shattered tv, her former-arm-now-pixelated-plasma-bones weirdly hanging in the air (she didn’t know what to do with it), she noticed the humankind lavender candle gracefully balancing on the heaps of shattered glass and plastic. The flame hissing and crackling like all the time. Forgetting that her hand has changed form or dimension, she wasn’t really sure and forgot it anyway, Sarah grabbed her humankind candle. It had no scratch and was intact.
The pixelated-plasma-bones remained pixelated-plasma-bones. Sarah’s eyes went from the sturdy TV build from the strongest and lightest steel to the fragile humankind candle in her hand. What should have survived her anger, was now a mess on the floor, while the candle, gifted to her by a stranger on a long train ride, survived.
The door bell rang. Sarah’s friend, label very loosely applied, was on the other side of the door. Sarah could open it and explain, or could she play dead she thought? The pixels on her hand enlarged themselves, pulsating like living beings.
“Oh my X, Sarah where are you? What happened to the lovely TV? Omg, omg, omg this is so terrible. Wait I need to video this” Susan kept muttering while tapping wildly at her glasses zapping past the start button several times. With her eyes fixated on instructions appearing on the inside of her glasses, she didn’t see the light: Susan’s contours became abstract. Before she was able to switch on the recording the humankind candle abstracted her existence away into nothingness’s.
Collect this post as an NFT.
Worldbuilding is a text based game (with a LLM) player I'm slowly building on Farcaster. The LLM is still bland and needs to be trained on hand-curated resources. Certain things can't be speed up. For debugging I had to write stories (it made debugging funner). Inspiration was @samantha candle. Story is added at the end. Enjoy the read (and try the bot) https://paragraph.com/@cheshirecat/043-worldbuilding
Lol! I love it! That is so cute
Omg I like how innovative you are 😍😍😍
Explore a new dimension of storytelling where imagination takes the lead! In a recent post, @kbc shares a vision of a co-creative experience using AI for storytelling. The focus shifts from speed to quality and fun, nurturing artistic innovation without saturating timelines with generic content.