Train people to use GPT-related tech to become superhuman team members. You can do it all yourself, sure, but that doesn't scale. You can either train GPT-tech to use GPT to scale or train humans. Right now, reliability is the most difficult issue, so humans are essentially to ensuring the results are correct.
Humans are useful for training future iterations of the closed source thinking model. Once the model is sophisticated enough to train itself humans will no longer be required.
