Someone said from 4.7k - 10k. 100k is a lot more than 10k.

And both are right!
Imagine the following situation: The output of an 74HCMOS gate is connected to an input of another 74HCMOS gate. Discussing this situation in terms of equivalent circuit, output is represented by a switch, which is closed either to 50Ohm resistance connected to Vcc or to 50Ohm resistance connected to ground. The input of following gate is very high ohmic, only about 5pF from input to ground are present.
Further assume, that another digital signal line is running very close to this connection between the two 74HCMOS gates. If these lines are very close to each other and no shielding ground plane was placed between them, a rather high coupling capacitance (Cstray) is present between them, which can make problems. Fortunately, there's always this rather low output impedance of 50Ohm present, which shunts stray coupled noise to power supply rails and finally to ground, if adequate power supply decoupling capacitors are mounted.
If the another signal line is carrying a 1MHz square wave and Cstray is 5pF, then this will cause an injected noise of some millivolts at input 74HCMOS gate. That's ok, nothing bad will happen.

But what, if the driving gate is not a 74HCMOS gate but the output of standard 'C51 microcontroller?
Then, there are situations, where this low ohmic output impedance is not present all the time. There are situations, where strong pull-ups are only activated for two oscillator periods. Afterwards only weak pull-ups are present, which cause the output impedance to increase heavily!

Assume, that finally 50kOhm are present. Would this high output impedance be able to shunt injected noise to power supply rails?
At 1MHz capacitance of 5pF shows an impedance of about 32kOhm, which is smaller than the above 50kOhm! So, output impedance of microcontroller is no longer able to relevantly shunt noise to power supply rails. Finally, an injected interference of several hundreds of millivolts must be expected at input of 74HCMOS gate. This is enough to seriously degrade signal integrity and can erode the timing!

If now an additional pull-up of 4.7k is present at this signal line, output impedance of microcontroller seems to decrease by a factor of ten, and by this also noise-injection is heavily decreased.
So, if your circuit suffers from heavy cross coupling, which can often be observed, when not using a solid ground plane, then pull-ups of 4.7k...10k are recommended. But, when there's no such heavy cross coupling, higher impedant pull-ups, like 100kOhm resistors will also work. Then, only input leakage currents must be considered, which can cause certain voltage drop.

Kai