Wednesday July 2nd 2025 and a sunny day in San Francisco. I found myself yet again on the 10:25 AM Caltrain to Palo Alto, restlessly checking my X feed. There's this strange energy in the air, as if something is about to happen. PA, my partner in crime, and I, released our vision for where we think AI can help human beings, and not in a hype-type of a way. We call it Supernova. Because space is cool, and have you seen the new image from Rubin?
The algorithm keeps telling me that by this point we should accept that our tweet releases (P's and mine) are past event horizon. I so wanna disagree, hence the blogpost. I love a good pun and PA loves the craft, and between the two of us, we covered a good ground on this magical land called X (formerly Twitter).
Okay, so by this point, you're probably wondering why you should care. Or, in our wild-ish dreams, support our vision and the threads. Don't get me wrong, I'm writing this in the hope you do. We wanna see it blow up, and I am not hiding my intentions, disregarding the advice of CO (yes, that's Claude Opus), who tells me to just trust the algorithm. Of course it'd say that, no? Cheeky thing.
Here's the story. PA and I met at a crossroads less than two years ago, December 2023. I was in Australia, chilling with kangaroos. Yep, you should see those beauties at sunrise, wondering why this strange human bothered their morning rituals. Tl;dr us two humans went through some stuff, the kind of stuff that makes you wake up in the middle of the night (and not for an ice cream trip!). It's gone, se fue, se fue, and a big part is due to timelines crossing on X.
The two of us decide to write a story. How hard can it be? Amin, PA's a computer scientist (like proper, RL stuff), and I am, well, an astronomer who likes to AI whisper. We didn't start with Supernova, we just wanted to understand how AI "thinks". Yup, a charged word. But here's the thing. We use the models, and we trust them with many of our daily life decisions. Some big decisions too. Many questions later, a question supernova-ed its way to us.
What if AI models had distinct 'personalities'—styles or tendencies—that directly influenced how they made decisions? We think we identified a way to tell, and we wrote a whole paper about it, which got accepted to the Actionable Interpretability Workshop at ICML 2025. My first ICML ever, heard there's some cool after parties too. Secretly (well not so much anymore), been eye-ing Anthropic's. Girl's gotta dream.
Back to research. What if, once we understand their personality, we could collaborate better with the models to spark connections, leave painful stories behind, come up with new stories? The kind of stories that lift you up, that say, 'it's cool, kid, pack your bags and go, there's a big world out there, and an even bigger universe'.
It all starts with a 'what if'. And if you don't believe me, I challenge you to come with a counterexample. The two of us are now daydreaming about the afterwards. Maybe it's a singularity, but we hope it's gonna be the stars - the big 'uns, the Wolf-Rayets of the world.
If you wanna help us write this story, we're gonna be chilling in the far out. Come check us out. And if you like this story, we'd love your support—likes, retweets, shares—in making our tweets (Biggie and Caddie) 24hr+ outliers.
--
A huge shoutout to the star PA, you know who you are, and you know where you're going. 'Uff', we did it.
Comments