I like tabular ML, but very skeptical about this research direction. Language models trained on a broad corpus make sense, as the language is similar across di

I like tabular ML, but very skeptical about this research direction. Language models trained on a broad corpus make sense, as the language is similar across different domains. Time-series, however, are extremely different... Stock prices, heart-rates, brain-waves, digital signals... Patterns learned on a broad dataset should only introduce uninterpretable noise.