To be fair, to some degree this is exactly the use case stochastic parrots (the thing we call ‘AI’ as a buzzword) can truly excel at:
Interpeting the bullshit we stammer out when we try to give a verbal command while totally not adhering to any reliable command structure.
Formulating a reply that sounds like fairly natural language despite how inane the sources used might be.
So yeah, we’re finally at a real use case. Gimme! And from briefly trying it, Gemini is better at figuring things out from impresice input than Assistant was.
Make no mistake, it’s ultimately the same backend. They just swapped the processing layer between audio-to-text parsing and running inputs from them (and again on the way back). Sadly no Google Now smartness at all, we’ve lost that forever. But hey, at least this improves stuff.