Is this the same idea as when we switched to multicore machines? The rate of change on the capabilities of a single agent has slowed enough now the only way for OpenAI to appearing to be making decent progress is to have many?