Directed Acyclic Graph
Battling the myth of more unexplored edges = no progress
With four weeks of “vacation” under my belt, I return to New York to start my sabbatical dazed and feeling more time scarcity. It’s confusing to try to fit a single coherent story for what I did the last four weeks. All of the following are true:
Visiting temples in Kyoto with Alice’s family, art museums in Seoul, savoring food we booked months in advance.
Talking to startups and AI labs in SF, where left and right I am told, “I feel an increasing heavy weight on my shoulders”; “every day matters”; “why haven’t you started interviewing?”
Skiing with my parents at Vail, sharing a bed with their Shiba, the first time I found peace looking into its eyes.
Making decisions for my wedding.
When did I feel most invigorated? It’s definitely when jumping between high energy people and ideas in SF. But as much as it felt exciting to debate and hear stories about AI and startups and progress, the conversations only highlighted the tearing tension between the part of me which feels “every minute matters” and the larger part of me holding back — whether out of rational skepticism or cognitive dissonance from not wanting to abandon the rest of my life. Choosing AI is kind of an all-in choice.
When did I feel happiest? Those quiet brief respites where I felt present and at peace: walking down Philosopher’s Path in Kyoto, sitting on the slopes in Vail and watching snow fall, admiring the beautiful language and drama of Scarlett O’Hara, connecting with family. Those were some moments of wonder.
Having absorbed an abundance of information which pulls me in many directions, I have to remind myself that the world doesn’t have my best interest at heart. Only I can decide what is important. I’ve had time to live, to play, think, to talk, to gather information, to think and talk and think more. But if my path ahead is a Directed Acyclic Graph (in fact it could be cyclic), I’ve only opened up more branches to go down. Seeing the amount of things to do, it’s easy to fool myself into believing no progress has happened, and that is a kind of all consuming stress.
So here’s my attempt to accurately document my state of mind, the nodes I’m currently in and the edges I see ahead.
AI
(Re)read pivotal papers, both to understand and relive the path of progress
attention is all you need, gpt2, llama, deepseek; alpha go → alpha zero, katago
Play around with models
I’m forcing myself to change habits and throw many queries at these chat interfaces. I’ve already spent more time talking to all these models the past few days than ~all of 2024 (not counting coding questions at work).
There are other explicit tests I plan to run
Go
it still struggles with coordinates and gives a mix of “ok understanding” and nonsense
I would be surprised if it cannot impress me in coding or math, given benchmarks
can it pass my interview questions?
I need to figure out how to evaluate Cursor / Cognition / coding in context abilities. I no longer plan to prioritize my RL project (which I could not get models to help me unstuck), so I need some other complex coding context to test.
Implement gpt-2/llama
This is both for interview prep, and probably offers good/necessary understanding
My current P(interview some set of companies by June) is 0.9 and P(working in AI in 2025) is 0.24 (*).
Personal
Go
I have set up weekly lessons and long games
I will try to go to 1+ meetups/week
I’ve identified some clear areas for improvement
It’s easy to sink time into this: I can just “play and review”. It’s less clear whether I’m making progress. Compared to a month ago, I’ve gained some clarity on directions of improvement, but also, I can’t clearly say I’ve grown in directions identified last year (**).
Exercise
I am finally setting up trials with personal trainers; for the past two years I’ve been satisfied with participation points but I recognize it’s not very efficient.
Need to figure out some consistent cardio also
Trips to see friends:
Boston
DC (& along the way)
Chicago
Bay Area/SF
It’s also easy to forget miscellaneous items which add up in time and energy consumption. To list a few:
relationship & friends & family
cooking
wedding logistics
reading
writing
In some sense, they are not the current priority; but in a sense, they are what truly matters.
I think it’s all too easy to connect the dots looking back, but the path is never so clear looking forward. I am both trying to figure out this whole AI thing and the balance between productivity and play — perhaps the ultimate goal is to reclaim what success means in my own terms. I hope my unabashed sharing resonates with someone navigating similar crossroads. I'd be delighted to hear what surprised you in my reflections & plan, or what parts of your own journey mirror mine.
*: 0.5 want to go * 0.7 get good offer * 0.8 nothing super exciting at JS * 0.95 coordinating with Alice * 0.9 unknown unknowns
**: I love losing a game like this from careless wishful thinking
