Live a life not in the training set

Live a life not in the training set

AI is trained by a supercomputer studying a collection of writing. This collection is called the training set.

State of the art AI models like ChatGPT have read every notable piece of writing ever written by every human in history. The foundation labs ran out of human writing a long time ago, so AI researchers are programming AI to generate synthetic writing to further educate AI models.

So is AI going to know everything and be the best decision maker ever? Not yet, because the training set lacks the rarest, most scarce, and most civilizationally impactful information. This data is tucked away in the minds of the few people who live out-of-distribution lives at the frontiers of human experience.

This type of data fits into two categories:

  1. non-linear scenarios and assumptions that are possible right now that were never contemplated by previous generations, and
  2. singular peak experiences that are not well-expressed or captured in written form.

For example, what does it feel like to be teleported into Donald Trump’s head right now? How do you think about leading America through a pivotal moment in history? How is he calculating on how to position America for the AI revolution, how is he calculating to maintain American hegemony, how is he calculating and trading off between different political, social, cultural, demographic, technological, and economic goals?

What’s it like being Elon Musk where he has the potential to be the most powerful and impactful human in human history by controlling global satellite internet, humanoids, self-driving cars, spaceships, solar power, and social media? What chess moves is he thinking about late at night? Should he try to just start his own country?

What’s it like being Sam Altman as he runs potentially the most important project in human history with OpenAI birthing self-recursive AGI/ASI? What is like to negotiate multi-billion dollar business deals with all major technology incumbents? How can you load and calculate an insanely ambitious multi-product, multi-sector roadmap? How do you raise $100 billion dollars in one shot?

These are three people in our era who are collecting singularly rare pieces of data that no entity has access to. You can't hire Mercor to find people to label these trajectories. This is the rare stuff that you have to work your entire life to be a position to even have a chance to play it out. What does it actually feel like to be the in chair to hit go on birthing super-intelligence? What’s it like playing strategic geopolitical poker with other thermonuclear superpowers and actually thinking about hitting the red button? What’s it like to actually dance around, bluff, and out-maneuver other super-billionaires? What’s it like to charm, dazzle, and actually close in ultra-high stakes negotiations that impact all of humanity?

Sure, there are confidants and biographers, but only these singular humans actually have full fidelity on every micro-detail of the negotiations, the human intuition, the subtle reads, the lessons from previous failures and successes, and those hard-won battle scars that are informing their wet, carbon-based neural networks on their decisions. Sam told me over dinner recently that AGI will automate his job away in ~2 years. He is too humble; the singular intuition and taste data in his brain is the hardest to replicate away. In fact, his job gets more and more important as AGI automates away all lower-level, solved execution threads and singular judgment is the only thing matters.

Their life data is not in the training set.

There are many people who take SATs and APs to get into college. There are many people that do AIME math competition and TopCoder contests. There are many people that have negotiated a condo purchase. There are many people that have hosted a tech podcast to debate if we’re in an AI bubble or not. There are many people that have bought and reviewed this Michelin-starred restaurant or that luxury car.

This data is in the training set. That’s why ChatGPT is so magical and can tell you exactly what to do across 99+% of human experience. A life path generatable and predictable by ChatGPT can be very satisfying, pleasurable, and meaningful, but to me feels aesthetically empty.

I need to live a life that’s not in the training set. I need to be out of distribution. I need to live a life and carve a path that no one else has tread. You should do it as well.

My heuristic to do so is simple: when confronted with a crossroad decision, choose the path that no one else would choose. Choose the path that gives you a crazier story. Choose the path that scares you the most. You’re living a video game, and choose the most extreme path that generates the wildest, weirdest narrative arc. All the normal paths are already in the training set. You can just ask ChatGPT for what the normie life is like. The only chance to escape the matrix is to keep choosing the weirder and weirder path until you jump off the map entirely.

I’ve been deliberate on executing this strategy. Building a new type of venture firm with Jake Paul and Logan Paul (they’re each running out of distribution life paths at an interesting convergence of culture and tech); investing in top AI and robotics researchers who are building the substrate of the future; creating a fourth macro-nutrient food category with Ketone-IQ, building marketing AI agents on top of social data with archive.com; these are all deliberate decisions to get me closer to the frontier.

Life’s too precious to live within the training set. Go do something weird.

Subscribe to Geoffrey Woo Thought

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
jamie@example.com
Subscribe