You Don't Know Who You Are
Fun with AI
These days I’m so deep into agentic programming that I need AI to tell me how much AI I use. Not only that, I’ve upgraded my CV precisely under some AI instructions to tell me exactly how much. Here’s the new page. By the way, you can see that with over 700 git repos to my credit, I’ve gotten around. So when people ask me if I’ve done xyz, I need to say something more specific than probably.
You may remember a product called Google Desktop. I used it to search everything on my desktop when I had many many gigabytes. It was discontinued in 2011 because Google saw the cloud scale terabyte writing on the wall. Now we have so many gazigabytes ‘we’ need datacenters just to run the compute for AI. The cool thing of course is that we are in the loss leader marketing stage of LLMs. This is like Best Buy selling you DVDs for cheap in 1997 just to get you into the store and buy a DVD player. I hate to remind you of what happened since then. First release DVDs used to cost 10 bucks and you owned it. You want to own a fresh movie now? Michael costs 25 bucks, 20 just to rent.
Anyway, my memory is twisted just like Simba’s. I don’t remember stuff in any form or fashion the way that Google or Anthropic’s old school or next generation bots do. So I went down the rabbit hole of building an Adventure Map. That’s because I dropped my daughter off at LAX this morning and she’s on her way to a merry Central Europe holiday. Vienna, Salzburg, Munich, Prague and Budapest. So I had to update my Been There Done That family document which is currently just a spreadsheet with no details outside of the number of times I’ve been wherever. So why not have Claudio interview me and get some flavor?
It turns out I do much better remembering my own life when I have an interrogator. Most of the time in casual conversation we have no idea how we misremember. For example, I remembered in human fashion that I saw Butterbean the 350 pound heavyweight boxer at the 96 Atlanta Games. Not so says my LLM. He was a professional at the time. Well, dagnabit I was there. Perhaps it was an exhibition bout at the Olympics, because there could be no other rhyme or reason for me to pay money to watch this dude I never heard of before.
This is the game of memory whack-a-mole I often find myself playing with AI. For example, when I talked about the trek I took from Horseshoe Meadow up to the peak of Mt Whitney in 1987, the number of miles I’ve always been saying is 57. Claude disagrees. But I really can’t tell where we camped the first night on the trail. Was it Crabtree Meadow? I’d accept the suggestion. Even thinking about it now, I can’t say for sure. I just remember something about marmots. Or do I?
I recall attending part of a morning mass in Milan, recognizing the exact part of the liturgy even though it was in Latin. I can’t tell you what year I ate dinner in a small restaurant behind the cathedral. I’m fairly sure it was between 2000 and 2001. That’s just about all I remember. This picture serves as a prompt to my memory. We grew up using photographs and postcards that way. Loss of a snapshot often entails loss of a memory.
Claude did help me find a song I remember from childhood. It was Red Bird. It was sung by Lead Belly, which I didn’t know even back then, I’ve been singing the wrong lyrics all along. Well that’s nothing new for any of us. But it couldn’t help me find the right recording of a song called (maybe) Crucifixion. I sure as hell could sing the part I remember “Not a word, not a word, not a word”. It wasn’t Marian Anderson, the singer was way more gut bucket. Not Ethel Waters. Not Sarah Vaughan. So the memory haunts me I suppose like it should. Like an invisible spirit, replaying in my mind without permission. Not like an exact digital forgettable I can sit in some terabyte somewhere.
What I’m discovering that to an even greater extent than before, there are practical reasons to trust computers with stories we tell ourselves, and a new nuance for and against the mysterious. Where does the mystery lie and what value do we place on fuzzy concepts? How valid is our own self-mythologizing? What is it worth? Where does accuracy matter and where does it not?
Working with LLMs in this manner are journeys into the plausible. It gives us a new way to think and a new way to remember.