This concept is a key part of my research.

I kept seeing these massive models for timeseries that were frankly mediocre at best. Using better preprocessing and smaller models (1k versus 1M parameters) and curated datasets I have met or exceeded similar works.

Smaller, curated models are the future IMO.