Tuesday, January 14, 2025

The frontend I’m working on creates an attractive web interface that lets me manage tasks from the web. Changes are pushed to the Markdown notes, and then automatically synced across all of my devices. Eventual consistency creates some challenges, but it’s been a great test case for my use of AI.

For the curious, I have a large directory of Markdown notes. In any note, I can add a task:

“ - [ ] Follow up with Lacey 📅 2025-01-13”

I have built some Python code that lets me query the directory for tasks, update them, complete them, etc. The directory is synced to all my devices.

My latest experiments have been geared toward writing full apps, rather than restricting the scope to smaller scope in order to avoid regressions. I’m currently working on a frontend to my text-based task management system. It’s been enlightening.

I have started to bump up against the limits of many LLMs for programming tasks. At a certain scale, responses degrade and often cause regressions. I have been impressed with o1, though. It's ability to "reason" by employing a "chain of thought" has significantly increased the ceiling.

Rams are cookin! Let’s go! 🐏 🏈