Fine-tuning InstructGPT or text-davinci-003

[wide knoll]

Is it possible to fine-tune GPT models that have been trained with reinforcement learning from human feedback, like InstructGPT, text-davinci-003, or text-davinci-002, or can you only fine-tune self-supervised learning GPT models like the original dvainci? Can you fine-tune Codex models?

[warm totem]

hi