A 35B open-weight model running on my machine beat Claude Opus 4.7 at a concrete task: drawing a pelican in ASCII art. Not an abstract benchmark. A real question about what we're paying for and how we measure intelligence.
I've been team Ollama since day one. Last week I tried replacing it with raw llama.cpp and a minimal wrapper. What I found forced me to rethink something I thought I'd already figured out: Ollama solves UX, not infrastructure.
I reproduced the tiny LLM experiment that blew up on Show HN: Gemma running in the browser, no API keys, inside my usual stack. Here's everything that broke — and the little that actually worked.
32 years in the dev trenches. Here I write what I learned, what I broke, and what nobody tells you in the tutorials.
No spam. Unsubscribe anytime.