Giving my smartest AI a veto, not a vote
I expected a smarter AI to fix my wildlife model’s mistakes. Instead it overruled 38 right answers and fixed one. Here’s the job it turned out to be great at.
Built by Matt
I’m Matt, a full-stack software engineer. This is where I keep the things I’m building, notes on what I’m learning, and a running record of what I’m up to.
A camera-first mobile app that names the wildlife you photograph and keeps a collection of what you’ve found. Before building the app, I’m testing whether the AI behind it can be trusted.
I expected a smarter AI to fix my wildlife model’s mistakes. Instead it overruled 38 right answers and fixed one. Here’s the job it turned out to be great at.
My wildlife app’s model was right 98.9% of the time and still couldn’t admit when it didn’t know. The fix was already sitting in results I’d saved.
Building a wildlife discovery app that answers “what kind of animal is that?” from my daily walks, and learning how to test AI properly along the way.
I’m Matt, a full-stack software engineer who likes building useful products, exploring new technology, and learning in the open.