What are reasoning over log probabilities?

Reasoning over log probabilities is like guessing which toy is more likely to be picked based on how much it "shouts" its chance.

Imagine you have a bag full of toys, some are big, some are small, and each one has a different chance of being chosen. Instead of just saying “this toy is 50% likely,” we use something called log probabilities, which helps us compare these chances more easily, like comparing the loudness of each toy’s shout.

Think of each toy's chance as a whisper or a shout, louder means it has a higher chance. But instead of measuring how loud they are directly, we use logarithms to make the math simpler. It’s like changing from counting steps to counting how many times you jump, easier for big numbers.

When comparing toys (or chances), using log probabilities is like having a better hearing aid. It makes it easier to tell which toy is more likely, even if the differences are tiny, just like how a small whisper can still be louder than another one nearby.

So instead of saying “this toy has 2 times more chance,” we might say its log probability is higher by a certain amount, and that helps us reason faster and better.