With gpt-3.5-turbo, would it be better to provide context as HTML or plain text?
What would give the best results? I’m thinking that HTML may have additional semantics line lists and tables that could help the model to better understand the content.
Thank you
All the extra < and > and other tokens might dilute your prompt, though.
Have you tested both? Best bet would be to run 5 or 10 tests both ways and note differences.
Welcome to the forum. Hope you stick around.
Well, I’ve actually asked and this is the response:
As an AI language model, I have been trained on a large dataset of text, which includes HTML markup. While I don’t have an understanding of HTML in the way that a human would, I am able to recognize and parse HTML code and extract information from it. I can also generate HTML code based on input data or generate output that includes HTML markup.
I will run several tests and see what works best. For me, it would make sense that data that is already structured in a table or a list to make more sense and be easier to understand.
We will see,
Thanks
Hi Mircea,
Did you perform your tests ?
Can you share your feedback?
Curious to know, as I am facing the same kind of questions!