🔥Fire.
Or: the feeling of unbridled exhilaration when holding an invention you know will change everything.
I feel like what our distant kin must have felt when they came upon fire. Or rather, were able to conjure it. Control it.
Pure magic. Power. Giddiness.
Minds racing with all of the possibilities. Brain cells exploding from overstimulation. From being asked to believe too much from what their eyes could plainly see.
If seeing is believing, then each day for the past 3 months has been an exercise in testing the limits of faith.
A mere 12 weeks ago I sat at my laptop and watched as a friend took me to the OpenAI playground and demonstrated what “it” could do - using twitter threads and blog copy.
Wow, I remember thinking. This would be great for 25 year old me - the me who was still a brand marketer trying to figure out how to write with a consistent brand voice and character across all the many touchpoint.
But today me? It felt like viewing cool technology as a bystander - impressive, but not for me. Not for the oh-so-human problems I was working on. Things like remembering which kid needed a violin and which kid needed library books today. Things like when swim registration was (dear god please get my kids into lessons so they may possess this basic life skill, but for the shortcomings of their parents unable to beat the crowds to the registration button) and getting those dates into our calendar.
Things like what I was making the kids for lunch next week.
If this really was so great, I could just tell it what I liked and it would tell ME what to make.
On a lark I decided to try it out.
Sitting at my kitchen table, I typed into the box: “My family doesn’t eat beef or pork. Give me next week’s school lunches that kids like to eat and are easy to pack.”
What I saw next is forever etched in my memory. “It” gave me 5 imperfect, but perfectly reasonable responses. Moreover, it understood the general shape of a school lunch included a main plus snacks and sides.
At that point I knew nothing about temperature or hallucinations, prompt design or few shot.
All I knew was I was holding fire.
Since that day I’ve learned a multitude of limitations but also a myriad of applications. Of ways that generative AI - or LLMs (Large Language Models) in particular - could be the very answer I’ve been searching for, for 3 years, maybe longer.
Of the ways that a technology whose intelligence is based on probability can be paired with humans whose intelligence is grounded in empathy to build things and augment humans in ways never before dreamed or deemed possible.
12 weeks later, I’m swimming in concepts and experiments - embeddings and vector databases. Logprobs and compute. RLHF and tokens. So many details that make building more effective and still, so flipping fun.
But no matter all the specifics, each morning I wake up thinking one thing:
Fire. I’m holding fire.
Beautifully written, Avni! So stoked to see what you've stumbled upon and how you're applying it to the problem space you're working on 🔥