Scale AI Hackathon
3rd place / San Francisco / Feb 2023
The weekend
In February 2023 I took 3rd place at Scale AI's hackathon in San Francisco. This was days before I started the YC W23 batch, so it doubled as my last unstructured weekend for a very long time. I spent it making language models rap.
The demo was an AI rap battle between Tom Brady and Gisele Bündchen, which was topical at the time and absurd in exactly the way early LLM demos were supposed to be. It was apparently absurd enough to make the local news, because the SF Standard wrote it up. I did not expect a weekend project to end up in a newspaper. In early 2023 that kind of thing just happened.
The era
It is hard to overstate how strange that stretch of San Francisco was. LLMs had just gotten good, nobody fully knew what they were for yet, and hackathons were the fastest way to find out. You would show up with a vague idea, wire a model into something by Sunday, and occasionally discover a real capability hiding under a joke. The rap battle was a joke. The fact that a model could hold two distinct voices and trade coherent verses between them was not.
Everyone in the room was doing some version of the same experiment: poke the model, see what falls out, ship the demo before the weekend ends.
What it turned into
That same energy is what became Helicone a couple of months later. Once you have watched a room full of people wire LLMs into everything over a single weekend, the question of who is going to watch all that traffic in production answers itself. The hackathon was the last project I built for fun before building the company. Third place, one newspaper article, and a pretty good weekend.