Personally I think we'll find something better than the LLM algorithm fairly soon, but it will still be using the same GPU type servers.