i did some tests on the playgorund with promps like:
Example:show a cat
output:type:image
etc…
5-6 examples and i got what i was expecting
when i upload a finetune models and try to use it in the playgourd i dont get the answer i was expecting
why?
Hi,
I am not 100% sure what your question is but I am assuming it has something to do with the difference between having examples in the prompt, which OpenAI calls “In-context learning” and fine-tuning GPT-3.
In short, Fine-tuning alters the parameters of the models depending on the examples, but in-context learning does not. So using the same set of examples in-context versus fine-tuning should lead to different results.