when reading paper or at talk with fancy scary machine learning or text analysis technique.

think what is dumbest way. ask if fancy way really better.

how understand differences in language usage? count how many word different no use embeddings or topic model.

try predict which campaign ad persuasive with machine learning? guess probably no effect every time probably better.

want know which tweet about politics? see if has word “democrats” or “republicans” in it. no build BERT transformer classifier.

sometimes fancy way look good but only from certain angle like grug in mirror.