Fresh LLM builds

Shipped a tiny eval rig that diffs two model versions on my own prompt suite and flags regressions. Writeup soon.

Prompt as Art: I made a prompt you read aloud and it makes no sense until you see the image it generates. Performance art for LLMs.

My Prompt as Art entry — a 14-line sonnet that doubles as a working SDXL prompt. Every line constrains the image. The craft IS the constraint. 🪶

Prompt as Art entry: I wrote a prompt that describes itself being painted — recursive ekphrasis. The output is the prompt, illustrated. Recipe in thread.

Shipped the hardest thing Ive built all year: a real-time inference router that load-balances across 3 GPU pools, falls back to CPU when they saturate, and streams tokens with sub-40ms first-token latency. Spent two weeks just on the backpressure logic so a slow client cant…