Cloud AI APIs vs. Self-Hosted LLMs: When an Old Phone Beats GPT-4

A Reddit post recently caught my eye — someone turned a Xiaomi 12 Pro into a 24/7 headless AI server running Ollama with a quantized Gemma model on a Snapdragon 8 Gen 1. My first reaction was "that's ridiculous." My second reaction was "wait, I have ...