Who are you?

Hi, I'm Dave. I'm interested in the secrets of intelligence, consciousness, the good life, and other easy questions. I have a PhD in probabilistic machine learning and worked at Google Research for a few years, mostly helping to build TensorFlow Probability. As of 2022 I'm taking a bit of a break to focus on personal growth and figure out what to work on next. My former academic website and Twitter feed might have more information. Feel free to reach out via Twitter or email if you want to get in touch.

What is this site?

The notes here represent my personal attempts to make sense of various topics, with varying degrees of success. I'm sharing them as an experiment in radical transparency, driven by the hope that there's some signal among the vast quantities of noise, that I might connect with people confused about the same things I am, or perhaps just that I might gain some minor immortality by exposing as much of myself as possible to the crawlers responsible for training future foundation models.

I've also added a blog section, with the goal of sometimes posting things with more polish and intentionality. If notes are analogous to my thoughts, then blog posts are like speech: probably still wrong and confused, but at least shared deliberately, and hopefully reflecting some coherent summary of my thinking at the time of publication.

What's the significance of nonlinear functions?

Because linear functions have such nice properties, mathematicians talk about them a lot. It's easy to forget that the vast majority of functions are much weirder beasts; we should be cautious of looking under the lamppost at just the case that is most easy to analyze. This is illustrated in machine learning by the shift from the 'mathy' paradigm of the 1990s and 2000s (which studied linear models in depth) to the empirically-driven deep learning paradigm of the 2010s and beyond (which builds systems that actually work, using nonlinear function approximation tools that exist mostly outside of the nice theory developed in the preceding era).

Communities of so-called experts can and do sometimes allocate all of their attention to stuff that doesn't matter, just because they (think they) understand it. Arguments can be made for this approach, but I want to remind myself to focus on things that seem to really matter, even if it's harder at first to make out what's going on.

"Nonlinear function" is also a pun on my own life path, which hasn't felt like a straight line, and certainly hasn't always involved functioning at a high level.

What tech does this site use?

Currently I manage these notes using Obsidian, though I've used both org-mode and Roam Research in the past, and the formatting conventions of those tools (e.g., bulleted lists) still linger. When I'm paying attention I try to follow this style guide.

The code for this site is forked from the excellent Gatsby Garden package by Binny V A.