Few-shot learning or fine tuning the chains of thoughts of o1 mini based on the CoT of o1 (new feature request)

hi! as of today, CoT from o1 models are hidden,
but could it be possible to, for instance,
run o1 on a few sample inputs, get the id of those sample api calls from the api, and then reference those sample calls to fine tune o1-mini’s CoT process, or to few shot o1-mini’s CoT (by showing it o1’s CoT) – but still in a zero-knowledge environment, if you want to keep your CoT secret, devs don’t need access to the CoT itself.
It would be amazing, and would, with very little effort, improve quality (and cost) by 10x, possibly more!
Thank you very much!