after weighing multiple factors including user experience, competitive advantage, and the option to pursue the chain of thought monitoring, we have decided not to show the raw chains of thought to users.
There must be some magic sauce here for guiding LLMs which boosts performance. They must think inspecting a reasonable number of chains would allow others to replicate it.
They call GPT 4 a model. But we don't know if it's really a system that builds in a ton of best practices and secret tactics: prompt expansion, guided CoT, etc. Dalle was transparent that it automated re-generating the prompts, adding missing details prior to generation. This and a lot more could all be running under the hood here.