I read this as basically finding a language representation of timeseries data that would be understood by LLMs better than feeding raw records. I'm guessing it

I read this as basically finding a language representation of timeseries data that would be understood by LLMs better than feeding raw records. I'm guessing it tokenizes very similarly to any other LLM. Perhaps I misread, though.