Two years on substack
Welcome to the 45th entry of my public journal, where I share *some of the things* I've been exploring across work, tech, wellness, and life.
Dear Friends,
December 3rd was the two-year anniversary of starting this substack.
Some thoughts:
By the numbers: 44 posts, 84 subscribers, 40-50% open rate, 100-150 views on a typical post.
My goal is to write twice a month. After 2 years I should have at least 48 posts. This post is number 45, so I didn’t quite reach my goal but it’s pretty darn close. I’m proud of my commitment and discipline. I have no goals for any metric except number of times I publish. I’m not trying to get more subscribers and I don’t care if people are reading these or not.
I’m happy you’re here. I just said I don’t care who reads this, and yet I’m still happy to share my writing with humans around the world who are taking the time to read it. Two things can be true: I’m glad you’re here. I would enjoy writing just as much if you weren’t.
Observation: When I first started writing I paid a lot more attention to my words. I remembered what I wrote about and when. As I continued to develop the habit of publishing every few weeks, I got less attached to the output. I don’t always remember what I’ve written about in recent posts. I enjoy the act of writing and care less about the product. I don’t worry about the quality of the output, and I no longer see it as a reflection on me.
Why I’m writing: Writing helps clarify my thinking and it just feels good. My whole body feels better after I sit down with my thoughts, write them in words, and hit publish. This is my motivation for continuing into year three. But I’ve recently realized another benefit: having a written record of my thoughts, observations, projects, and interests could be incredibly valuable in the future, especially as we increasingly rely on LLMs and AI. For example, I’m in a group chat with my second cousins, and we’ve been trying to piece together stories about how our great-grandparents immigrated to America. Fortunately, we can still call some of our parents who’ve kept written records or remember a few stories—but it’s mostly just small fragments. Imagine having access to a chatbot that could tell us about their lives in vivid detail.
Have you tried adding a new habit?
What was it, how did it go, and are you keeping up with it?
Farcaster AI Bot Updates
Last week I wrote about building an AI Agent and how I got out an MVP of a Farcaster bot that will answer basic questions about Farcaster if you mention it in a cast. I continued to work on it this week, adding several new capabilities:
Added memory so the bot can have a back-and-forth conversation with someone, simulating a natural chat, in case the user wants to ask follow up questions. It was more difficult than I expected it to be. I eventually realized I was approaching it wrong and ended up writing a recursive function to rebuild the chat history starting with the current cast and building up, instead of starting with the first cast in a thread and going down.
Infinite Chat. It didn’t take long for someone to introduce my bot to another bot account, thus kicking off an infinite chat between two bots since both were designed to reply whenever they are replied to. When bots are designed to have the last word, the conversation never ends. It’s not an ideal scenario when you’re paying for LLM usage, so I ended up taking my bot offline while I figured out what to do about it. I decided to temporarily hard code a limit to the length of conversation my bot will have. There are some better alternatives, including using AI and letting the bot decide when to stop a conversation, which I may revisit later.
Gave it more knowledge by creating embeddings for the remaining Farcaster 101 videos as well as 5 videos from the Here for the Art series. I updated the prompt so the bot also knows what knowledge sources it has.
Gave it an ETH and SOL wallet so it can participate in onchain activity in the future, but the bot doesn’t know about it yet so there’s nothing happening there yet.
Improved its responses. The transcripts I trained the bot on weren’t perfect. For example, the bot thought my cohost was Nannish Prav instead of Nounish Prof. “Garbage in, garbage out”. I could have updated my prompt and told the AI the correct name, but I wanted to correct it at its source so I went back and did a search and replace of all the transcripts I’ve loaded, and reloaded them into the vector DB.
Try it: If you’re on Farcaster, feel free to test it out by tagging the account @gmfc101 and ask it a question!
What’s Next: I could spend all day adding features to try to catch up with some of the other bots on Farcaster: Aether, with its onchain capabilities and conversational skills; Clanker, for its token deployments; AskGina, with its comprehensive knowledge base; or mfergpt, for its awareness of what people are casting about. But I’ve decided to focus instead on what makes my bot unique. I can always go back and improve other areas later, but for now, it’s better to lean into what I can do that others can’t. That means giving it access to the many hours of GM Farcaster video content we’ve created and improving its responses based on this knowledge. So far, it’s been trained on just 17 videos—but we have over 190 episodes it hasn’t accessed yet. My focus will stay on loading it up with this knowledge and maybe iterating on prompts. Let’s see where we end up!
Have I mentioned how much fun I’m having?
Some of the (other) things:
A rundown of some interesting new onchain projects, including a mention of CastOut, a survivor-like game we’re launching for GM Farcaster.
The Hub, a developer focused Farcaster podcast launched recently with host Dylsteck.eth and already has 6 episodes, as part of the gmfarcaster network.
We’ve had some great guests on GM Farcaster recently. Dune CTO Mats, Farcaster cofounder Dan Romero, Faust founder July, Clanker founder proxystudio.eth, Supercast founder woj.eth, Stoic and voteforpedro from Solana, Eulerlagrange.eth and dawufi from Opacity talking zkTLS.
Episode 14 of my (just for fun) podcast with 3 of my old college friends is out. The topic was Manifest, and you can find it wherever you listen to podcasts.
And last, I’m an animated gif! Breck went to a nvidia hackathon recently and starting building a streaming platform for Farcaster called Togger.com; he added ability to take a clip and turn it into an animated gif and he tested it out last week when I was on a live stream with the one and only Samantha who makes hand made candles at humankind.place. Buy a candle and tell her I sent you!
A Note to My Readers
I started this substack in December 2022 as an experiment to see if developing a writing habit would help clarify my thinking and/or provide other benefits. You can read about my original intentions in my first post or my more recent reflections after sticking with it for a year.
I write about twice a month and share musings, meditations, and links to things I’m finding interesting as I build out a farcaster-native media company, a modern technology consulting company, raise my kids, and have fun creating and learning in the worlds of crypto, tech, finance, science and wellness.
Thank you for supporting my writing and journey. If you’d like to get in touch you can reply to me here or find me on X and farcaster.
Until next time, keep putting good into the world. —adrienne🌏❤️