Reaching a plateau doesn't imply that it's not early. It's still entirely plausible that we come up with a newer better model in ten years that gives us true AGI or runs on cheaper hardware or just gives us a closer approximation to human reasoning.