I never said that you couldn't predict with a sample. I said that it is mathematically impossible to accurately predict an outcome with a sample that is less than 1/2 of 1% of the population. There is a difference between statistics and mathematics.
Plug in 1 for confidence interval. That means you want the result to be +/- 1% margin of error.
Check 99% for confidence level. That means that the you want the sample's result to fall within the specified confidence level 99% of the time. I.E, if you repeated the sample 100 times, 99 times you'd be within that +/- 1% of the true number.
Enter 225000000 for population, which is approximately the eligible voting population of the US.
You will see that the sample size needed for this level of confidence is only 16,640.
This is only .007% of the population.
These numbers aren't some crazy magic they invented just now. This math was first developed like 75 years ago, and involves some intense calculus, and you can double-check it with experiments and computer simulations.