Every age mistakes its machines for something new. The loom, the engine, the model. Each arrived as a rupture, and each was the same river, only faster. What holds is the bed it runs in. Truths settled before fire, certain to outlast the last circuit.
You steer by what the current cannot move. Humboldt crossed half the earth not to collect its variety but to hear the single grammar beneath it. Marcus Aurelius held the reins of the largest machine of his age and never confused the empire he commanded for the order that commanded him.
I have gone looking for what they found. Through capital markets and machine learning. The length of Europe to the Bosphorus, the American West, the cloud forests of Costa Rica. I was never after the countries or the fields, but what held beneath them all.
The intelligence now rising will move faster than anything before it. Steered blind, it raises a Babel. Anchored to the truths that have anchored us, it raises us. They are not its brake but its engine, and finding them is the work I have chosen.
This is the most important crossing our species is now attempting. I work to steer it.
Prior exploration
A cross-attention mechanism that infers robot actions from world-model state transitions, retrieving the closest demonstrations from memory. Cuts action-prediction error 6.5x over a direct MLP baseline on LIBERO-Spatial using just 25 demos, then fine-tuned with GRPO reinforcement learning.
A decoder-only transformer built from first principles: custom byte-pair tokenizer, causal self-attention with masking, residual blocks, and an autoregressive engine with temperature and nucleus sampling. Eight heads. Seven million weights. One question at a time.
DQN and PPO agents trained through self-play on Connect Four, built on a CNN-ViT hybrid backbone where convolutional layers scan local threats and a Vision Transformer attends globally across the board. PPO reached 98% win rate, DQN 85%.
Equities