The author of this paper runs the inference team at OpenAI. So I would imagine that this is a record of her experiments without necessarily revealing which ones they're actually using in production.