Newcomb's empty box

I have written about Newcomb’s Problem before. Because the world is small, you keep stumbling over the same topics. So a couple of days ago I watched a Computerphile video about LLMs and Newcomb’s Problem. They use a box of $50 that may or may not be filled depending on whether the predictor, let’s call it oracle, thinks you’ll also open the other box with a certain $5.

For some reason, in my mind the choice was always between a certain $50 and an uncertain $5 or $55 for opening both boxes. But of course, if you don’t have faith in the oracle’s accuracy, then you may end up with $0 if you pick the one box but it thought you would open both.

I suppose that that isn’t a big problem in the classic scenario where the oracle is known to be very accurate, but it helps me understand better why people would consider to two-box, I think.