[arram]

Teaching is used in the colloquial way, not in a technical ML sense. The model isn't being retrained or fine tuned.

[typon]

The prompt is essentially fine-tuning with extra steps

[minimaxir]

Not really. Fine-tuning fundamentally changes the model weights to be more amendable to a particular use case/domain; the few-shot prompts for GPT-3 is just a guide (and it's very easy for the model to ignore said guides and go on a tangent).

If you could finetune the 175B model, you'd likely get even better results for these Q&A prompts. (unclear how the OpenAI API is implementing its finetuning demo but I believe it's not on the 175B model).

[IshKebab]

I think they only fine tune the top layers.