"Working with a cathode ray tube, for example, we can manipulate the voltage on the plates and directly observe the deflection of the point of light on the screen. We can use the equations from a physics textbook, combined with voltage measurements on a multimeter, to make predictions about the electron beam’s deflection and then test them directly on the device."
I would argue a small proportion of humans have verified the physics they've learned as you describe. The rest learn from textbooks and ultimately there is an element of trust that they have that most of the books are not false. Ultimately from a Bayesian perspective, I think it is possible to obtain very high confidence about certain attributes of the world even if your sources are very contaminated with fictions.
Anyone who majors in physics will have performed lots of experiments in the lab, including this one. The average person may believe what they read in a high school science textbook but their justification rests on the word of others and the general institution of education. Occasionally, we have seen situations where this institution is under attack [1]. Nevertheless, I would not consider something to be knowledge if you only read it in a book. Heck, even experiments can lead people astray [2] as later experiments may contradict earlier ones.
All of this is to say that it is very hard to know anything with a high degree of confidence. The confidence many people have is a false one. However, in the case of LLMs, there is no basis for any confidence whatsoever. The text is the claim and LLMs have no means of independently verifying that claim. All they can do is statistically evaluate the internal consistency of the text, and even that is fraught with difficulty. "A Lie Can Travel Halfway Around the World While the Truth Is Putting On Its Shoes."
I would argue a small proportion of humans have verified the physics they've learned as you describe. The rest learn from textbooks and ultimately there is an element of trust that they have that most of the books are not false. Ultimately from a Bayesian perspective, I think it is possible to obtain very high confidence about certain attributes of the world even if your sources are very contaminated with fictions.