I built an LLM RL Training Pipeline

I LoRA fine-tuned then RL'd Gemma 1B to play the game Takeover in a multi-stage training process. Takeover is a competitive card game with a punishing action economy where your goal is to rid your hand of all your cards as quickly as possible. Takeover is

I Have Been Unemployed for One Year

I left Google one year ago as part of a voluntary layoff package. I joined as an L3 Software Engineer and left as an L5 Senior Software Engineer. I worked hard and was very good at my job but I wanted to experience more of the world. In this past

Concepts That Make Administrative Work Simpler

In your job, you are either delivering your core work product or you are doing administrative meta-work that makes the final delivery easier. If you are an event photographer, your core work is being there; taking the photos. Your meta-work includes scheduling with clients and tracking shoots you

Multipolar Powers

I live in a world that is wholly in my mind and anyone can visit if they bring a song to accompany them. These songs represent the visitor and these songs are the visitors and these songs form the multipolar powers that decide my fate. Through these powers I am

198 Days Post-Google

I took a voluntary layoff from Google ~200 days ago. I have split my time between software, literature, and governance. Previous articles: One Week of Unemployment Two Weeks of Unemployment One Month Out 136 Days of Unemployment These first 6 months were part sabbatical, part work. I am officially back

Note on Ender's Game-Era Capitalism and Portfolio Management

Suppose we have reached endgame capitalism. Suppose we have colonized the worlds as in Ender's Game. How should we manage our stock portfolio? Investing in many uncorrelated assets de-risks your position. In 2026, there are not enough sufficiently uncorrelated assets for complete de-risking. In Ender'

How Government, Society, and Freedom Relate

The opposite of freedom is oppression. Man is born into a state of freedom. Through a man’s interactions with the world, he finds himself oppressed or restricted in ways that limit this natural freedom. Man ought to introduce government in accordance with the principle that a society is only

First View on sprites.dev from fly.io

Yesterday I read Code and Let Live, an article covering fly.io's new(ish) Sprites offering. Sprites (sprites.dev) offers persistent VMs that feel like ephemeral VMs—offering ~1s latency to fresh shells, state snapshotting, and a CLI that makes remote code execution a breeze. I created an

Alembic Migrations with SQLModel Common Issue Resolutions

"I have a database whose schema doesn't match my SQLModel models. I need to migrate this. I want to use alembic's – autogenerate flag. I am running into issues" If this is you, that was me 15 minutes ago. Here is a Q&A

136 Days of Unemployment

I keep a Google Doc titled "Notes". I look at it every day. I feel guilty whenever I look at it. I feel guilty because "Notes" has my TODO list. Notes has a log of how slowly I DO my TODOs, if I DO them at

So, You Want To Build a Road

If you are reading this, you are considering building a brand new road! First, congratulations! Building a new road is a serious undertaking and you probably don't even know where to start. That's okay! This instructional will walk you through your first "roadie" considerations.

What I See Coming in LLM Reinforcement Learning

RL for LLMs is going to be a relevant and durable process for AI development so long as we don't have AGI. Reinforcement Learning (RL) for LLMs seems to be a popular topic right now for those of us in the AI space; however, most of the world