At this point, most conceivable beneficial use cases for LLMs have been covered. If the economics of AI tech were aligned with making a good product that people want and/or need, we'd basically take everything we have at this point and make it lighter, smaller, and faster. I doubt that's what will happen.