Daybook, 16 July 2026

Working from home, and the loop that wouldn't cooperate. Most of the day went into evaluating my agents and working on the optimization loop — the one that can rewrite some of the skills, or the prompts, plus harness optimization. But it just wasn't working that day. So I did the honest fallback and read — some small papers, and there's a good blog specifically on harness optimization.

Finding my writers. Somewhere in there I had a long session with Perplexity, trying to figure out the authors and people I actually resonate with. And over a series of conversations, three came out: Dostoevsky, Tolstoy, and Camus — the one who wrote The Stranger. That's steering my reading toward more memoir, personal experience, the connecting kind of book — deliberately away from the shelf I already own.

Because here's the honest self-assessment. The nonfiction has done its job — the logical side of me, the way I think about technology, data science, society, organizations, business — the math / business / tech / science side — is well grounded. What I've been feeding for the last three months is the other side: connecting books, like The Correspondent, Bird by Bird, Theo of Golden, and Mother Mary Comes to Me. Books that make me think more empathetically. And honestly, the conversations over these months have shown me the same thing — that I'm actually pretty good at connecting with people, talking to them, understanding a situation. Which is why I still can't quite accept that someone once called me a narcissist — it reads as the exact opposite of who I've come to know myself to be. There's the old Kural line for it: the wound burnt by fire heals from within, but the scar left by hurtful words lasts a long time.

And the counter-move I've settled on — fewer people, but deeper. Maybe six conversations a week at most, and each one a real, aligned exchange, not just a once-in-a-while “hi” or some not-so-fruitful chat. It keeps my bandwidth clean — to read, to explore, and to have the conversations that actually go somewhere.

Organizing the research. On my own side, I'm pivoting — organizing myself toward five areas, and testing how well I can actually produce good research artifacts in each: epistemology; AI safety and alignment; vision-language and multimodal models; agentic assistance; and generative / causal and probabilistic modeling. The target is concrete: before the first week of September, before I head back to school, I want a really good set of things done — from the internship and from my personal research. And I've learned my own grain here — for some of this, it's better to work on it alone first, and then find the right institution or lab to collaborate with. That's what I find effective.

A deadline of a different kind. I ordered Tolstoy's The Death of Ivan Ilyich and Homer's Odyssey. I want to finish the Odyssey before August 11 — I've booked a ticket to watch it on a 70mm IMAX screen. I think I'll be the first person in my family to watch a movie on a 70mm IMAX screen. Read the thing first, then live it.

Settled · 16 Jul 2026