OpenAI's secret is that it is building toward the singularity
If you ask any artificial intelligence researcher what OpenAI’s secret (in the parlance of Peter Thiel’s zero-to-one) is, they’re likely to say that OpenAI figured out the scaling hypothesis early and built towards scaling AI systems.
I would take the view instead that based on the company’s early experience solving games and simulated environments, their stated goals to build a household robot, and their partnership with Microsoft, that OpenAI believes they will usher in the singularity.
Specifically, they believe they will usher in a machine-learning platform fine-tuned on human augmented-reality training data and made to run on robots, first simulated robots, then physical robots, then simulated robots again.
In my view (and based on what they have shipped), the current focus on the GPT platform is an external face that is meant to usher us slowly into the age of machine intelligence and an internal focus on quick return-on-compute.