Manual feeding data Vs. training the model with data

What is difference between manual feeding the data to gpt and ask her to learn it for a specific session Vs. Training the model itself with data?? (Say some data of 2023 for which model has not been learnt yet)

Hey there and welcome to the community!

By “manually feeding the data”, I’m assuming you mean “provide information to GPT in a conversation”?

So, if we were to break this down in a simple way, let’s understand that AI is a big bundle of algorithms and math formulas.

When you talk to an AI, you think of it like “running” the algorithms, making it whir basically. Nothing changes to the algorithm when you use it. When the conversation gets bigger, you are simply increasing the size of the variable you’re putting into the algorithms.

When you train the model, or fine-tune it, you are changing the algorithms in a particular way. Training/fine-tuning these models are methods to change what are known as model weights. These weights, while we don’t precisely understand how they work, are what affects its output. So if you feel as if you need to tweak the actual algorithm to optimize it for a particular task, that is something fine-tuning is suited well for.

This is just a basic introduction, but this should still help you in understanding what’s going on :slightly_smiling_face:. I hope this clarifies some things for you!