Alibaba Cloud Computing Reinvented Reinforcement Learning

I love leverage, the fulcrum that lets you move the world with a small push. I look for leverage across all areas of my life.

First, it was using Python to find the hidden alpha in financial statements back when I worked on Wall Street.

Now, it’s about building and backing companies that embed AI into their DNA. I’ve seen firsthand that the raw power of machine learning is staggering, but I’ve also seen its Achilles’ heel: it’s an expensive, inefficient, and often brittle learner.

We’ve been training our most advanced AIs with the subtlety of a firehose, drowning them in data and hoping they learn to swim. It works, but it’s a colossal waste of energy and potential. We’ve been building digital calculators when we should be nurturing digital intellects.

That’s why, every so often, a research paper lands that feels less like an incremental step and more like a paradigm shift. It offers a new kind of leverage. The paper on VCRL, or Variance-based Curriculum Reinforcement Learning, is one of…