Denny Britz

My name is Denny Britz. I am interested in a variety of topics around automation, artificial intelligence, and the human mind. I enjoy being a beginner - learning complex topics from first principles and discovering simple, intuitive, and beautiful ideas.

I have worked as an Artificial Intelligence researcher (Google Brain, Stanford University), HFT-style algorithmic trader, distributed systems and database engineer (UC Berkeley, Apache Spark), and have been in involved with a variety of early-stage startups as a co-founder or software engineer. I used to write at WildML, a popular blog during the early stages of the Deep Learning era. I enjoy working remotely and living in various places around the world, mostly in Asia. I never completed High School, dropping out in ninth grade because I spent a bit too much time playing video games.

The easiest way to get in touch with me is via Twitter or email.

Places I’ve lived in

  • 🇩🇪 Germany (18 years)
  • 🇺🇸 LA & Bay Area (8 years)
  • 🇯🇵 Tokyo (3.5 years)
  • 🇹🇭 Bangkok (1.5 years)
  • 🇰🇷 Seoul (<1 year)
  • 🇸🇬 Singapore (<1 year)
  • 🇧🇷 Brazil (<1 year)

Current Interests

Artificial Intelligence

Artificial Intelligence and Machine Learning have been a constant interest throughout my life. I got hooked with writing bots for video games in junior high school. I later worked on distributing Machine Learning algorithms on top of Apache Spark as a research engineer in the AMPLab, NLP for Knowledge Base Construction (KBC) using factor graphs (a type of probabilistic graphical model) as a graduate student at Stanford University, and Deep Learning for NLP and Reinforcement Learning as a resident at Google Brain.

My interests tend to be guided by Occam’s Razor - I believe in simple and intuitive models that have an element beauty to them. I am generally skeptical of human-engineered complexity.

A few of my current interests are:

  • Algorithms inspired by nature. Using insights from biology, physics, information theory, and other scientific fields to build intelligent machines.
  • Meta-learning and un/self-supervised learning: Techniques to learn good representations without encoding inductive biases into model architectures and without providing explicit rewards.
  • Learning from simulations or mental models and transferring knowledge to the real world

Consciousness and the human mind

Over the past few centuries we humans have done an amazing job at understanding the physical world around us. But I don’t think we have done a particularly good job at aligning the rapid technological progress with our biological needs, or the needs of other organism on the planet. Some of our technologies are economically successful only because they exploit our own evolutionary programming or the environment around us.

We have barely scratched the surface of understanding our own bodies and minds. I think the next centuries will be about that, and designing a world that is conducive to our evolutionary needs and makes us happy. In addition to the basic sciences and artificial intelligence, I believe that spiritual practices like meditation and psychedelic substances may play an important role in understanding our own minds.

Writing

If you can’t explain it simply, you don’t understand it well enough. ~Albert Einstein

Putting thoughts into writing is the most effective tool for deepening my own understanding of a topic. It’s surprisingly easy to fool oneself into an illusion of understanding through sloppy reasoning, especially if the conclusions confirm one’s prior beliefs. By consciously trying to avoid the curse of knowledge and expressing ideas in simple terms, you can get around some of that. If anything on this blog happens to be useful to the reader, that’s amazing, but that’s almost never my primary goal. I am mostly writing to learn myself, and that’s why I don’t usually write on topics I consider myself an expert on (as if there were any).

Open Science, academic incentives, and reproducibility

While I’m not really working in academia, I have had several run-ins with the academic system in Artificial Intelligence as a graduate student and researcher. Many scientific fields are suffering from a replication crisis, but that’s not all. Coming from the industry, I naively assumed that science was all about discovering facts. I was surprised to find that academia is also full of misaligned incentives, and that climbing the academic ladder can be a complex, unfairly biased, and political game.

I am interested in tools and mechanisms that make science more transparent, efficient, and accessible to everyone. The open science movement, which includes open-source code, open peer reviews, and open-access distributions servers like arXiv, has already had a big positive impact on some scientific fields (like Machine Learning), but I believe we can do much more.

Algorithmic Trading

I see markets as being at the intersection of human psychology, gaming, gambling, automation, and artificial intelligence. I spent a period of time working on HFT-style algorithmic trading systems and loved it. Because they are results-driven and not dictated by politics, the markets proved to be an ideal environment for the scientific method and offered a fast feedback loop. I am quite interested in applying some of the latest advances in Artificial Intelligence and automation to that field.

Gaming

When I was in junior high school, I competitively played Counter-Strike (beta to 1.6) for one of the top-ranked German teams. My dream at that time was to become a professional eSports player. Making bots for Diablo 2 and Ragnarok Online is what got me into programming. But that only lasted until World of Warcraft was released and took over my life :)

Some of my favorite games of the last decade include NieR:Automata (perhaps the best game I’ve ever played without being influenced by nostalgia), Persona 5, BotW, Yakuza 0, and FF14 Shadowbringers. I also enjoy watching Twitch streams in my free time.