Transformer models have clearly demonstrated that you can convert anything into an input embedding and the AI can learn from it, even if the embeddings are from

Transformer models have clearly demonstrated that you can convert anything into an input embedding and the AI can learn from it, even if the embeddings are from drastically distant domains.