> where is GPT 5 going to scrape the next set of answers for new libraries and frameworks..?
The bottom line is that it doesn't matter as long as you have a large enough sample to learn the format, which is already there with existing data. There isn't an SO answer for everything anyone needs even about current tech, and the reason models can still answer well in novel cases is because they can generalize based on the actual source and implementations of the frameworks and libraries they were trained on.
So really, you only need to add the docs and source of any new framework during pretraining and bob's your uncle, cause the model already knows what to do with it.
The bottom line is that it doesn't matter as long as you have a large enough sample to learn the format, which is already there with existing data. There isn't an SO answer for everything anyone needs even about current tech, and the reason models can still answer well in novel cases is because they can generalize based on the actual source and implementations of the frameworks and libraries they were trained on.
So really, you only need to add the docs and source of any new framework during pretraining and bob's your uncle, cause the model already knows what to do with it.