We have a prompt for our RAG pipeline which asks the GPT-4 model to respond in the language of the user query. So if the user asks a question in Spanish we ask the model to provide a response to the question in Spanish.
When doing this I’m noticing an issue that occurs reasonably frequently :
So for characters like ó the model sometimes responds with a backslash before the utf-8 representation starts. So for the word Colección the string that should be sent back is
Colecci\u00f3n, but what we end up getting back is Colecci\u00f3n.
Is anyone else observing something similar?