I don’t fully understand how GPT operates and would appreciate some help.
I’ve noticed that through contextual training or repeatedly focusing on a specific type of question, such as grammar issues, GPT’s accuracy improves significantly in a single session.
However, once the window is closed, it requires retraining.
Can I use GPT for specialized training and then interact with it on this specialized topic via an API?
What I mean is, can I interact with a GPT that has undergone specialized training and shows improved performance in a specific domain through the API? Is this currently possible?
What steps are necessary? If there are any articles on this topic, please recommend them. Thank you very much.
When you use the chat completion API, you can load in as many examples as you like (up to the token limit).
What you can also perhaps take a look at is fine-tuning, if you will be using this method a lot. Fine-tuned models are more expensive, but the tradeoff might be worth it if you have to load a lot of context every time.
If you want to stay within the chatgpt ecosystem (e.g. with a custom GPT) - you might consider simply loading the “training” conversation into your instructions/system prompt. that might be enough to get you started.