Mostly my experience, but “Gemini crunches data” would be my replacement there.

If I have a task that requires parsing through swathes of irregular data that traditional ml would choke on (or require an intermediate training step ala bigquery), I have gotten much better results from Gemini than the other two.