Training costs might be coming down but costs for hardware that can run these models is still obscenely high and rising. We're still nowhere near a point where its realistically feasible to run a home LLM that doesn't feel like it's suffering with severe brain damage.