I think that's the funniest part. Like, as far as I know, the regular Assistant uses the same approach to handling data that buzzword AI things use, a neural network. But branding (and potentially internal company politics) is weird, so they decided to kneecap Assistant in order to make Gemini look better on release.

Hi Bob!

YMYDYMYD

I kinda thought the title made it clear I was an American.

The way statistical sampling works, 1000 people in a population of 300,000,000 is actually good enough for most things. You can play around with numbers here to convince yourself, but at 95% confidence 1000 people will give an answer to within 3% of the true answer for the 300,000,000 population.

Check out Ecosia. It's the same thing, but your searches plant trees in a responsible manner.

Suddenly trying to convince all my friends and family I'm from France.

Hi, Bob!

Lol, I'm sorry you're getting downvoted for speculating about improving weights and measures in a thread about wanting better weights and measures.

I mean, yeah, 1000 people is enough assuming there's no sampling bias. But if you've got sampling bias, increasing the sampling size won't actually help you. The issue you're talking about is unrelated to how many people you talk to.

Your own suggestion of splitting up the respondents by state would itself introduce sampling bias, way over sampling low population states and way under sampling high population states. The survey was interested in the opinions of the nation as a whole, so arbitrary binning by states would be a big mistake. You want your sampling procedure to have equal change of returning a response from any random person in the nation. With a sample size of 1000, you're not going to have much random-induced bias for one location or another, aside from population density, which is fine because the survey is about USA people and not people in sub-USA locations.

DST is good actually. Fite me.