Custom GPTs can be a point in the middle of fine-tuning and few-shot prompting, since they allow to apply Retrieval Augmented Generation (RAG) to expand the answer of the model with files from a knowledge base.
Going through the human imitation path, the knowledge base could basically contain the human gameplay logs. They don't necessarily have to be format in a specific way as in fine tuning. In principle, I understand that as long you include how the data can be used by the model in the instructions text, it should be able to use it.
You can create, play and share custom GPTs easily in the ChatGPT web-app. However, to use them as an API model endpoint, you have to create them first in the playground, and they'll appear then as a new model in the playground chat drop-down with a model ID that you can plug in the OpenAI Python API.
Custom GPTs can be a point in the middle of fine-tuning and few-shot prompting, since they allow to apply Retrieval Augmented Generation (RAG) to expand the answer of the model with files from a knowledge base.
Going through the human imitation path, the knowledge base could basically contain the human gameplay logs. They don't necessarily have to be format in a specific way as in fine tuning. In principle, I understand that as long you include how the data can be used by the model in the instructions text, it should be able to use it.
You can create, play and share custom GPTs easily in the ChatGPT web-app. However, to use them as an API model endpoint, you have to create them first in the playground, and they'll appear then as a new model in the playground chat drop-down with a model ID that you can plug in the OpenAI Python API.