To be precise, the claim was that when you can improve an algorithm by randomizing it, you can improve it further by derandomizing it.

Yep, I figured that out eventually.

I'm leaning toward the notion that they get an initial idea in their heads of what someone else seems to be saying, and then discard all data showing that the person is saying something else - confirmation bias to the point where it destroys the ability to read.

This may be the case for the instances you're thinking of, but doesn't seem to cover all instances if poor reading comprehension. I regularly converse with someone whose comprehension difficulties appear to stem from a poor short-term memory, for example.

This seems to be another point in favor of having a less-meta subreddit or forum - we could share resources about that kind of peripheral skill there.