What caused psychologists to be so wrong about the size of working memory?

In the TED talk he's making a joke ("memory used to be X, now it's Y!"). Likely referencing the "good ol' days" trope where one talks about how things used to be different when the comedian/speaker/audience was younger, or more specifically the tendency of memory to decrease with age.

The "span" of working memory depends on the type of "object" being remembered, how exactly it is remembered, and what other distractions or tasks are interfering. The original "7 +/- 2" paper is quite old, and although it did propose a limited working memory capacity with a particular span, I do not think it supports your claim that psychologists "used to think that the size of working memory was 7 items". Rather, in one experiment with one type of object the average was 7 items. People had to do further experiments with other types of items to find out if the span was constant over all objects; once doing those experiments, they found that it was not, but consistently found that there was some capacity beyond which it was difficult to remember more objects without more elaborate strategies or training (such as the "memory palace"/method of loci).

Wikipedia has a pretty extensive article on memory span that is probably a good place to start if you're interested in the topic. Some of the characteristics listed on that page that influence the span include relatedness of the items, distractors, rhythm and rate of how objects are presented, modality (auditory, visual, etc), motor cost (longer words take more "space"), how exactly responses are scored (total recalled? number before error? how are adjacent/similar objects scored?), and other distractions.