2 ms·
> converts time-series data into text-based inputs an LLM can process. What? Which the model then tokenizes? I am struggling to make sense of this.
by beardyw 2y ago
> converts time-series data into text-based inputs an LLM can process.
What? Which the model then tokenizes? I am struggling to make sense of this.
- ubercore 2y agoI 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.
- beryilma 2y agoThis is the modern day equivalent of "every signal is an image" after the initial success of deep learning algorithms in image classification tasks. It is just the academia chasing the hype train...