Roni Kobrosly Ph.D.'s Website

Not a full blog post, just a quick update on my first robotics project. The TLDR of it: I tried to make a robot car for my daughter, it's been challenging and I made the mistake of going maximalist, but it did make her excited. This was a fun and humbling experience, and I have a lot of respect for engineers living at the intersection of hardware and software...

New blog post wherein I try to learn Rust the old-fashioned way (books and exercises) and with an agent that has one hand tied behind its back. The Rust compiler and borrow checker thrashed me REAL GOOD, but eventually it clicked. Also, a defense of learning a programming language in the age of agents. Working slowly, struggling, and gaining technical "muscle" (all while getting constrained help from AI) is worth defending.

I came across a really cool open source project while scrolling through HackerNews the other day: auto-identity-remove. It automates the painfully tedious process of opting out of data brokers and people-search sites.

Life is so large and overwhelming and fluffy and indeterminable. These are tough times. I find a lot of comfort in spending some of my free time studying math, logic, and engineering. It feels like I'm reading the secret language of the universe, and it's crisp, internally-consistent, and explainable ones and zeroes...

To be totally honest with you, I have no fucking clue what is going on with the cost of AI inference. I hear partially-informed murmurs almost every day. Is it getting cheaper because of improvements in this new field of "Inference Engineering" or in GPUs? Something about TPUs? Or are our complex reasoning models and larger context windows resulting in greater costs? Maybe the improvements and greater token throughput balances out in the end? 😅 What does it mean, precisely, when we say "AI costs are subsidized"?

My father has an old picture of my grandfather that he loves. Unfortunately it's blurry and artifacted. For the last five years, at the end of each year, I've taken a few hours to see if cutting edge tech tools for digital image restoration could improve it. They failed miserably until EOY 2025...

Let's say you've fully instrumented your organization's applications with OpenTelemetry. You're capturing tons of useful metrics, trace data, and service attributes. Congratulations! Will incident time to recovery drop? Let's think through this before you celebrate too much...