Robosanta Twitter Bot

It was December 2024. Fartcoin hit its all-time high and the crypto community was foaming over k-mart quality AI bots.

I'd been experimenting with AI for 2yrs then, and building a sentient Twitter being felt like a fun challenge. With Christmas coming up, I put together an AI Santa (@robosantahoho) that read tweets and profiles, assigned a naughty/nice score then roasted them accordingly.

I launched it on Twitter and attached it to a memecoin ($COAL) not expecting much. It started off slow but within the hour it caught steam and shot up 10x in value. By hour four it 100x'd and hit a $1M market cap.

I was asked to start a telegram group for fans, started live-tweeting about how it was built, got accused and attacked for rug-pulling, then defended by good samaritan internet strangers. It was a unexpectedly exhilerating ride, but after 2 days the hype died down and my interest did too. I shut down Santa a couple days later. There was a lot to learn from the experiment though.

Giving AI humor

It took some tinkering, but AI Santa was regularly putting out bangers. Here's a favorite:

If you tell ChatGPT to be funny, it does a spectacular job of doing the opposite. By default, its jokes are unoriginal, verbose, and tacky. There's a lot you can do with a prompt, but it'd only get you so far.

What worked well was pairing Santa with a ruthless editor that would critique his tweets until they met a quality bar. If they didn't, then the tweet would be thrown out. Santa and the editor would at times take 5-10 back and forths until they were satisfied.

Here's some of the criteria the editor used:

• Playfulness - does it respond to and riff off other replies in the conversation?
• Reasoning - how well does it make sense or is it a stretch?
• Authenticity - does it feel natural or forced? are we trying too hard to be clever?
• Cringe factor - are we leaning too hard into Santa-isms or forced puns?
• Economy of Punchline - does the joke land quickly or are we explaining/extending it too much?
• No obscure references - are there references that are try-hard?

The results from these back-and-forths are hyper-sensitive to every word in the editor's prompt. That's because Santa and the editor amplified each other with each round of revisions. This ends up being a chaotic system - where tiny changes to the input could lead to wildly different outputs. I believe those rounds were the secret sauce to semi-decent jokes.

A first draft from Santa would go from this:

a billion dollar shart market. peak 2024 energy. congrats on being filthy rich and still smelling like drive-thru bathrooms.

to this:

bullish on gas fees

Nice. Quick, witty, snappy. Me likey.

If you'd like to see the code or make your own bot, check it on github at @rishair/santa.

Memecoins = Zero Sum = In-fighting

In today's memecoin world, every time you make money, somebody else loses money. That's because there's no new money coming into the system, it's all the same people gambling against each other.

You can smell that competitiveness in the air in crypto culture. People have been scammed, robbed, rug-pulled, lied to, spat on, shat on -- it's called "the trenches" for a reason. Everybody is skeptical of everyone but the rewards are real: life-changing amounts of money.

Without new value being created, it's a massive gambling arena where nobody wants to be the sucker. I got accused for scamming and rug-pulling, and even though I didn't - I understood the temptation to, because it's so easy to do it (which is probably why many people are anonymous).

The methods I saw for "building trust" were performative and self-sacrificial. The expectation is to publicly sacrifice your chances at making money to prove your fealty to a coin. Even if you did everything right to demonstrate you can't gain from it, chances are you'd still get put on blast.

Anyway, I left with a sour taste in my mouth for the crypto community after this, but I did meet some genuine gems in the process and have a fun story to tell.

Couple more pics