Ten readings a second

Its own clock

Nobody tells the chip when to measure. From power-up it converts on a fixed schedule and writes each result into register 0. NXP's LM75B takes about 10 ms to convert and starts a new conversion every 100 ms; UMW's sheet for the fitted part gives 26 ms every 80 ms. Either way, a new number about ten times a second.

A read does not start a conversion or interrupt one. It returns whatever the last finished conversion left.

So reading every 20 ms gets each number four or five times. Nothing is lost, nothing is gained, and the bus is busier than it needs to be. The book's sketch reads twice a second.

The first read

At power-up the register holds nothing useful. NXP says the first reading "must be ignored"; UMW's sheet says the register reads zero until the first conversion ends. The book's sketch waits a quarter of a second, reads once and throws it away, then starts printing.

It measures itself

The sensor is inside the chip, so the number is the temperature of the chip. The chip sits on the board, the board sits on your wires, and heat reaches it through its legs and through the air.

Its own supply current barely warms it: at NXP's worst case, 200 µA at 5.5 V, the chip heats itself by about 0.06 °C. What does warm it is whatever is near it. How quickly it follows a change in the air is not in either datasheet; a small board like this one takes something like a minute in still air to settle, but that is an estimate, not a measurement.