Noob AI q: Is it possible / likely that various LLMs will be able to land on some interoperable underlying data format so their training can be additive with one another rather than from scratch? Not talking about fine-tuning.