I sold my property to go all-in on global asset allocation. 🌏
Managing a complex portfolio across multiple currencies and brokers was a nightmare. Firefly III is great, but I wanted something faster, modern, and built with Go.
So I built my own.
Introducing GAAP: A high-performance, self-hosted personal finance platform. 🚀
The funniest thing about China’s AI boom:
The only people consistently making money from AI are not AI startups.
It’s the people selling AI courses to people who want to start AI startups.
The Book of Genesis
And the programmer said, "Let there be a repository," and Codex initialized git. This was the first minute.
And the programmer said, "Let there be structure — let the entry points be separated from the implementation," and so src, tests, and docs were each set in their place. This was the second minute.
And the programmer said, "Let there be interfaces, that others may know how to use this," and types and contracts were made manifest. This was the third minute.
And the programmer said, "Let there be tests, to divide the working from the broken," and the red grew fewer, and the green grew many. This was the fourth minute.
And in the evening, as he reviewed the code alongside his thirteen subagents, Codex spoke unto him:
"5-hour limit reached. Resets at 9pm."
And he rested.
Running Qwen3-27B (UD) with Codex — handles single-file edits pretty well 👍
One annoyance though: can't disable thinking mode, and the chain-of-thought just drags on forever. The wait is real 😮💨
Unpopular framing:
TikTok wasn't just a clever app. It was the product of a deliberate state policy — China aggressively cut mobile data prices and built out 4G, then its content ecosystem filled the bandwidth.
The rest of the world is now scrambling to upgrade networks to handle the traffic that policy created.
Infrastructure policy as soft power. Unintentional, but effective.
Microsoft built its first in-house reasoning model without distilling from third-party outputs—and says it now stands alongside Claude Opus 4.6 on SWE-Bench Pro. Impressive on paper. But benchmarks are claims; widespread developer use is evidence.
MAI-Thinking-1 is our first in-house reasoning model developed from scratch that is competitive with models of similar size on STEM reasoning and coding tasks.
35B active/1T total MOE.
💻Coding: 52.8% on SWE Bench Pro competitive with Opus 4.6
🧐 Reasoning: 97% on AIME 25
🤝Preferred to Sonnet 4.6 on blind side-by-side tests