How much of a difference is there in the answers from a base gpt3 / davinci model vs a finetuned model? (I'm assuming this finetuning isn't just the addition of

How much of a difference is there in the answers from a base gpt3 / davinci model vs a finetuned model? (I'm assuming this finetuning isn't just the addition of a few examples in the prompt)