Rendered at 20:16:15 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
jncfhnb 2 hours ago [-]
This whole synthetic sampling thing really boggles my mind. It’s like… it’s difficult to come up with an example that doesn’t feel like a hyperbolic strawman. It’s just that dumb.
0xy 2 hours ago [-]
AI polls are fake, real polls are fake, and Nate Silver's modeling is REALLY fake.
Nate Silver implied Florida was in play in 2024 for Democrats, then it went +13 R. This is after he spent 9 years clinging to polling which systematically undercounted Republican support due to either sampling bias or shy voters, or were simply outright fraud in other cases (Selzer's Iowa poll).
jncfhnb 2 hours ago [-]
That’s just fundamentally not a meaningful argument
You need to assess calibration. Not “accuracy”. What does “in play” even mean?
arc_light 2 hours ago [-]
The issue is that AI doesn't understand human emotion or the way a crowd reacts to a sudden political shift. In the middle of a campaign, people change their minds based on the moment, and no matter how much data you feed it, a machine is always going to be a step behind when things actually start to heat up.
jncfhnb 2 hours ago [-]
No not really. It’s just if you ask the model what percent of people like apples, it doesn’t know the answer. We don’t even need to get to the level of change of opinions.
Models don’t “poll” their training data. If you give it five yes answers and five no answers, it doesn’t think that it’s 50/50.
If you give it actual data it can derive it, but that would be a poll
Nate Silver implied Florida was in play in 2024 for Democrats, then it went +13 R. This is after he spent 9 years clinging to polling which systematically undercounted Republican support due to either sampling bias or shy voters, or were simply outright fraud in other cases (Selzer's Iowa poll).
You need to assess calibration. Not “accuracy”. What does “in play” even mean?
Models don’t “poll” their training data. If you give it five yes answers and five no answers, it doesn’t think that it’s 50/50.
If you give it actual data it can derive it, but that would be a poll