Sorry, enlighten me - why is pair programming such an awful idea, in your opinion?

You are paying 2 people for what is, 95% of the time, less than twice the output of one person.

There are particular mentoring objectives that you can hit with it, and there is value in bouncing ideas off another person, but I don't need to sit in front of a computer screen for the latter.

There are definitely benefits that are larger than just mentoring or bouncing ideas. Often, the sum is actually greater than the individual parts, as fewer mistakes get made and more fruitful paths get explored earlier/faster.

Most coding mistakes are not very expensive.

As the other poster said, most coding mistakes are not expensive and the ones that are should have automation/testing in place around them. It is definitely not worth having someone look over your shoulder and eat up the little breaks you naturally give yourself. There is a reason that hardly anyone at companies like Google, Amazon, etc. do pair programming.

My situation is a bit different.

I often go on remote calls with people and our task is to implement something while also training the person or giving tips on how to do it in the future while answering any questions they have in real time (related or not).

So it's more of a 1 on 1 teaching gig for an hour or 2. Not 8 hours of pair programming every day full time.

Pair programming is great. But as with all things, only in moderation. An hour or two per week can actually be quite enlightening and eye opening.

What exactly is empirically wrong with Pair Programming?

Then what is empirically wrong with trio programming? If the pair programming wonks actually have data to show that it is a net benefit, then they surely must have evidence showing that 2 is a good maximum on return.

Like to this day, I have never seen any reproducible strong evidence provided that pair programming is beneficial. And I don't know if I would call the occasional ad-hoc (unstructured) "pairing" anything other than what a normal person would call it: "two people collaborating like normal fucking human beings."

But if there is evidence that 2 is better than 1, why is 3 or 4 not even better?

Have you personally tried it? It's great!

So call it collaborative programming. People tend to work well with someone checking their work, and more than two is limited by human communication. You're creating a strawman that I'm saying "The more, the merrier". I'm not.

So you think that pair programming is never a good idea?