The laser has an load impedance of 10 kΩ. Before reading this I had built a single poll RC low pass with R = 1kΩ and C = 44 µF, which would give f = 1/(2piRC) ≈3.6 Hz. The 10 kΩ loading reduces this frequency down a bit and divides the voltage down a bit more as well, but not by much. I chose this because the response of the slow input is on the order of 1 second. But maybe a slightly higher cutoff with more poles might be a better approch.
I had trouble with the PID loop with this single pole LP filter. Basically when I activated the script the proportional gain would overshoot and then keep accruing offset in the same direction. This happened in steps, I assume this has to do with the 1 second update time of the PID script. I have no intuition as to why it would do this, I tried positive and negative values of proportional gain (with integrator and differentiation terms turned off). I'm guessing the LP filter introduces some large phase delay? Or I've screwed up in setting the pole frequency way too low compaired to the 1 second loop time on the PID script?
Will the internal thermal pole of the laser unit no filter the high frequency junk on its own?
Depends on the series resistance of the slow input. Maybe that's in the manual? The response of the SLOW input can be measured by driving this input and looking at the FAST voltage while the cavity is locked. Its roughly 1 GHz/V for the SLOW input and 5 MHz/V for the FAST, so the SLOW should only be driven by a ~1 mV signal. There is a thermal constant which makes the FAST/SLOW crossover stable, but you should put in a low pass filter with ~2 poles below 10 Hz to remove the high frequency junk getting in there.