Hi OpenAI staff,
I’ve been exploring your fine‑tuning framework and noticed there’s limited public detail on the methods implemented. Is it full model fine‑tuning, layer‑specific tweaks, or some other variants? Can we get more details on this?
I’m particularly concerned that full fine‑tuning could lead to catastrophic forgetting, so I need to understand and assess that risk before deploying the model.