Blog
Thoughts, mostly about technology and philosophy. Sometimes about cats.
Summaries might be LLM generated. Actually, they probably are. Yes, they are definitely all generated by LLM. They're normally okay but I'm not going to take responsiblity for them or anything.
-
Gemini Pro Can't Patch
But neither can most LLMs, and it's kind of a hard problem if you think about it.
-
Playing Doom in 2025, or, Sculpture from Clay
Middle-aged man admires the clouds.
-
The Unreasonable Effectiveness of Fuzzing for Porting Programs
Symbol-by-symbol fuzzing provides an unexpectedly robust approach to porting C programs to Rust.
-
Cleaning Cast Iron
The joy of cast iron pans.
-
Reflections on Sudoku, Or the Impossibility of Systematizing Thought
Reflecting on the Entscheidungsproblem, I consider why a single method for solving all problems is impossible, using the Sudoku solver saga as an example. True problem-solving, I argue, isn't about one system but about building a diverse mental toolbox and knowing how to apply its contents.
-
On Stepping Back
It's easy to get so invested in a task you lose sight of your original goal, something I recently learned while porting C code with an LLM. I've found that regular, structured breaks at natural time boundaries help you step back and re-evaluate if you're still on the right path.
-
Web Development: The Last Bastion of Online Human Expertise
Web development is likely one of the last jobs AI can't take because its dependencies constantly shift and break, leaving humans debugging things like why Tailwind CSS suddenly ignored their config.
-
2025 03 20 Highways Our Original Sin
Reading about the housing theory of everything got me thinking about the unexpected role highways play in urban development. Highways allow residents to access services from workers commuting from afar, removing a self-interested push to support local density and transit.
-
I Wish Elon Musk had More Imagination
It seems like a strange use of time for powerful people to focus on firing employees instead of tackling more interesting problems. Even wildly speculative ideas for using vast wealth and influence feel more productive than simple layoffs.
-
Stargate SG-1 was the Last Great Sunday Television Show
Stargate SG-1 was the perfect example of the "Sunday Television Show", a now-rare type of light, episodic TV ideal for lazy afternoons. This format, built for casual viewing, seems to be fading in the era of streaming.
-
On Unproductive Work and the US Economy
Many US jobs focus on managing zero-sum games within systems like healthcare billing or law, which creates little new societal value compared to productive work. These complex dynamics are difficult to change at a systemic level.
-
On Variance as an Explanation for Dominance in Certain Fields
Pondering US electrocution statistics, I wondered if differences between groups might be more about the spread of a trait (variance) than just the average. It's interesting how a little bit more variance can create a lot more outliers at the edges.
-
Anki/Jisho Chrome Extension
Building a Chrome extension to look up text on Jisho and create Anki cards was remarkably easy using an LLM; it even handled adding Netflix subtitle support with minimal cleanup.
-
LLM Generated Flashcards
Memorizing vocabulary is a challenge in language learning. I built a small tool using LLMs to generate flashcards or Anki decks with context sentences and translations to help.
-
Noteomatic - Beautiful Understandable Handwritten Notes
I built a tool called Notematic to convert my handwritten journal pages into digital HTML or Markdown using LLMs, finally merging my physical and digital notes. Now I can easily integrate scanned notes into my other digital workflows.