The principle is about who does the work. Telling hands over a verdict, "he was exhausted", and asks for agreement. Showing supplies the evidence and lets the audience arrive at the verdict themselves, which they trust far more because they reached it. The engagement comes from the small act of inference.
It is over-applied. Some information is pure admin (a date, a location, who somebody is), and dramatising it wastes a scene on a caption’s job. The rule is worth following exactly where an audience’s belief matters and worth ignoring where only their comprehension does.
In visual work the failure has a distinctive shape: a voiceover stating what the picture is already showing. Two channels carrying identical information is one of them wasted, and the one you can hear is usually the one to cut.
What does "show, don’t tell" mean in practice?
Give the audience evidence and withhold the conclusion, so they perform the inference themselves. A judgement they reach is one they believe; a judgement handed to them is one they assess.
When should you tell?
When the information is administrative (a date, a place, a name), and only comprehension is at stake. Dramatising a caption wastes a scene.