Today I screwed up on something I should have known better on. Like the rest of the many early adopters, I have automated agentic workflows plumbed all over.. N8N, launchd, etc. "Let's save some API money," I said to myself. So off with my deepinfra.com api key with an account loaded with $50, I tried both open weight LLMs and SLMs to do what an LLM did daily in my daily comms summarizing workflow. HALLUCINATION NATION. Though points for Gemma - the latest 31B is actually close to LLM-level on a long text summarization task. But there's so much hype around open models, even the people who work in AI research, myself included, get FOMO that maybe we can save some pennies on our household automations. Nope. You still get what you pay for: SLMs and open weight LLM models by non-frontier labs are *not* production ready (yet). They are at least 18-24 months behind the leaders..