Writing.

Five months running TunerBench. Time flies. It feels like it’s been forever and no time simultaneously. I truly love what I do. They say if you love what you do, you never work a day in your life. I’m not sure about that! I feel like I live and breathe TunerBench.

Everyone in tech is scared right now. So I went to the primary sources — BLS occupational reports, JOLTS, OES wages, and QWI — and traced software and related occupations over time. A longitudinal look at what the data actually says.

It’s always so lovely to be back in NYC and reconnect with all the talented and ambitious people here. For the last three years I’ve been living full-time in San Francisco and very heads-down on startups. Now that TunerBench is in stable alpha, we finally have some time to meet with different communities.

Nearly five years ago, I helped organize Brooklyn College’s first-ever hackathon. It’s been incredibly cool to watch new student organizers carry it forward and keep building something bigger than any one year.

When people say fine-tuning, they usually mean: change the model’s behavior by training it on a curated dataset. There are two common ways to do that.

A catch-up for anyone behind on the last 10+ years of AI — structured as Q&A.

I fine-tuned for the first time. It took all week, multiple platforms, and a lot of new jargon. Fine-tuning is a great way to shape behavior and style — consistency of tone and persona across prompts, more reliable reasoning, less prompt glue, more predictable outputs at scale.