Or maybe it's a combination of all of the above? I'm just guessing because i'm not a photographer, but it seems logical to me. At any rate, I guess we'll see what the results are when we see real world images from the camera.
You could have a 500 megapixel sensor, but if your sensor is not sensitive to light, then that resolution will be useless. Or you could have a 5 megapixel sensor that's ultra-super-duper sensitive to light, and get wonderful night shots.
But yes, the improved night photography is PROBABLY a combination of improved AI, an infrared sensor, and a sensor that's very sensitive to light – add photos at multiple exposures to the mix, and then combining them and using the AI to denoise them.