It could be added to context only on the inference server, with the model heavily instructed not to repeat its contents. Not foolproof, obviously.
It could be added to context only on the inference server, with the model heavily instructed not to repeat its contents. Not foolproof, obviously.