Why Empathy and Evals

I can't believe how many of you asked me this. Genuinely didn't see that coming.

So here's the story.

A few years ago, my friend Dash asked me a simple question: “What’s your superpower?” I had no idea. I stumbled through the conversation, and it wasn’t until near the end that something just came out. It landed somewhere at the intersection of empathy and conceptualization.

I’ve always learned through doing. Through trade, through work, through figuring things out over time. That’s the only lens I know how to look through the world. And it’s not something I just believe in the abstract. I’ve spoken and written about it before, including on Aditi Surana’s podcast and in a TED talk Dash and I did together on deconstructing empathy.

As the world evolves, work evolves, and roles evolve. I’ve been thinking a lot about what it would mean for someone with that kind of lens, sitting at the intersection of empathy and conceptualization, to come and help shape AI. Not just from a product or engineering angle, but from the place that asks: what does the end user, the customer, the agent actually need?

That question kept pulling me toward something I’ve been learning more about over the past few months: evals.

I don’t want to get into the semantics of it here, where it came from, what it technically is. What I do want to dig into is the philosophical intersection between these two words. Empathy finds its way to the heart of a person. Evals guide the nondeterministic nature of AI toward building for that heart. I think of evals as a guide, because what they really do is establish what the expectations from a given scenario could be, or should be. And then, with a little work with judges, you can actually get there. You knock it out of the park, basically.

Quite simple when you say it like that, right?

So I took two words I deeply relate to, found a domain, and decided to show up here and post what I learn, as I’m learning it.

That’s it. :)