Notes
On AI agents, data infrastructure and developer tools. RSS feed
- LLM output token limit: truncated JSON and a silent fallback
· A 41-finding scan outgrew the model's output limit and quietly switched providers. Splitting the work fixed it, and splitting it too far broke something else.
- How a falling-sand simulation works, with a playable demo
· Grainfall is a small C++ falling-sand sandbox. Here's how sand, water, smoke and fire come out of a grid of cells and a few local rules, and you can play it right here.
- Stopping an LLM from making up numbers in an AI analyst
· How I stopped an LLM marketing analyst from writing numbers it didn't compute, and the five places they still slipped through.