Vault Newsletter content is collected from Patrick’s digital notes activity over the course of the month – drafted by an AI agent and edited by Patrick.
September’s notes went into the final chapters of Visakan Veerasamy’s Introspect and a secondary book summary it pointed him to, Typographical Number Theory in Gödel, Escher, Bach, a map of the factions around the AI industry, and more of Kevin Davey’s book on trading systems. September was the month of Patrick’s wedding celebration.
Introspect, and where it pointed
Patrick brought his Kindle highlights from the last part of Visakan Veerasamy’s Introspect into his main note on the book and wrote his reflections alongside them. These chapters are about facing the parts of yourself you’ve disowned.
He doesn’t take all of it. On the book’s claim that blame only discharges pain and makes things worse, he splits the claim in two:
I’ll consider it true that self-blame is not helpful. But blame is a useful tool society wields against its defectors
The book quotes Solzhenitsyn’s line that “the line dividing good and evil cuts through the heart of every human being.” His reply:
But.. In reality its more like there really are evil people! Therapy mindset goes too far, steps outside its lane
What he did take: a passage about giving the world what you needed most in your darkest hours; he found it compelling.
He crossed off several of the book’s recommendations as he acted on them. Two new books went onto his reading list, and a Forte Labs summary of The Body Keeps the Score got its own annotated section in his notes. His verdict on the summary as a whole:
I’m finding this book review kind of scattered. … It kind of feels like a “throw all the new-psych terms together”. I think this methodology, presented this way, is not quite right for me.
Gödel, Escher, Bach, continued
Patrick’s GEB notes reached Typographical Number Theory, Hofstadter’s attempt to write arithmetic as a formal system, and a gap in it. Every individual sum can be derived, and the general rule still can’t. The system proves that 0+0 is 0, that 0+1 is 1, that 0+2 is 2, and so on down a pyramid of instances, but it can’t prove that 0+a equals a for every a. His note:
It can’t be derived. Unless we add another rule, a “rule of all” to let us derive that. Otherwise, we could say “~Aa:(0+a)=a” and even though it sounds wrong, it doesn’t violate any other axioms of TNT. Just like how Euclid’s 5th postulate can be violated and, while unintuitive, the result doesn’t violate the other postulates.
Mapping the AI factions
Patrick added a network diagram of the factions around AI, built from an article by Michel Justen of the Tarbell Center for AI Journalism. It starts with a lineage: the Extropians mailing list, Eliezer Yudkowsky and LessWrong, and Yudkowsky introducing Demis Hassabis and Shane Legg to Peter Thiel, whose investment helped found DeepMind. From there it branches into the effective-altruism and Open Philanthropy orbit and its ties to OpenAI and Anthropic, then sets the other camps around it: ethics and bias, e/acc, the Tech Right, the populist-right opposition to the Tech Right, and “AI as normal technology.”
Beating the monkeys
Patrick continued Kevin Davey’s Building Winning Algorithmic Trading Systems into chapter 12, on limited testing. This got into the specifics of testing strategies:
Once you test on a body of data, that data is burned. Every subsequent test is by definition an optimization.
So Davey tests on as little data as will give a statistical signal, then puts a strategy through a series of gates. The last gate is the “monkey test.” Run 8,000 random versions of the strategy: random entries with the real exits, random exits with the real entries, and both random. The real strategy has to beat about 90% of them on both net profit and maximum drawdown, in all three tests. That test design is one of the core nuggets of Kevin Davey’s whole system.
Also, the testing doesn’t stop when a strategy passes. Davey’s example is a system that beat 99% of the monkeys in development and, over 18 months, slipped in performance until it was no longer outperforming them. That was his signal to stop trading it.
Making things
Patrick finished the sculpture he’d been sourcing vendors for in August: Love Across the World, made for the wedding. It began as a cake topper and outgrew the job. A globe structure holds silhouettes of him and his wife on opposite sides, and branches grow from each silhouette and meet inside the sphere to form a heart. He modeled it in Blender, had it 3D-printed, and sent it to a Maryland shop to electroplate it in nickel. It stood at the reception.
A day earlier he published the wedding post: copying over the story of how they met in Bangkok, the years of travel between Asia and New York, and the proposal on a beach. It included the two ceremony readings, one of which was the Ortega y Gasset passage he had been working through in August: “The type of human being which we prefer reveals the contours of our heart…”
The first dance got its last fixes before the day: go earlier into the carousel, let go of her right hand on the spin-outs, bend forward to draw her in before the box step.
Living
Patrick and Pari had gotten married at the Manhattan courthouse in January, but the celebration with family and friends came in the middle of this month, with people there from every part of his life.
Honeymoon planning carried on through the final week, and Patrick dug up old notes to refresh his context about northern Italy: notes on a Leonardo da Vinci biography (plus generating a Mermaid timeline to track the places da Vinci lived), notes on European political history, and notes on art history. He also reviewed his Italian language notes as part of his attempt to brush up on the language before the trip.
Stray notes
A rant on what he calls the un-diversification of thought:
The correlation in the marketplace of ideas. Everybody is repeating the same news/tweets/tiktoks. The celebrity, your barber, and your friend. They all believe the same thing and state it with the same exact words because they’re just repeating. They’re not weighing it against their own experiences, because they have none.
An idea for a custom Thai translator app that listens for Thai speech, shows him a live translation, and plays his English replies back in Thai in a synthetic copy of his own voice. Then the broader point:
This is a new approach to software in the age of AI. Just vibe code what you need to do the job, and discard it afterwards.
A line he copied out from Eliezer Yudkowsky: “If you want to sound deep, you can never say anything that is more than a single step of inferential distance away from your listener’s current mental state.”
A quote from the film Seven Samurai: “Kids work harder than adults, but only if you treat them like adults.”
