I've used optimizers that run on neural networks and genetic algorithms, each with varying degrees of effectiveness, but I can't imagine an LLM being particularly useful, except maybe in writing and parsing documentation.