1) Autofocus. Check out Sony A7 III or A9 reviews, it's absolutely mindblowing. It focuses perfectly and is insanely fast.
2) Low-light sensitivity. You can get a used Sony A7s, which goes to ISO 490,000, and quite usable up to ISO 50K. At its highest ISO and with a relatively cheap f1.4 lens it's much more sensitive than my eyes - it can see in the dark, it's like night vision, but in color.
But that's what's available for purchase. Canon has demonstrated Canon ME20F-SH, which shoots up to ISO 4.5 million. Judging by the videos, it's absolutely nuts. I hope that tech becomes widely available soon.
3) The cost of RAW video cameras has plummeted. Blackmagic BMPCC 4K is $1300. Quite an amazing little camera that records straight to an external SSD via USB-C. This is cinema-level camera, though with quite weak ergonomics. Still, if you're serious about cinematography, it's an excellent choice, and it's very affordable.
4) Lenses keep improving as well. New 135mm/f1.8 is breaking MTF chart records, it outresolves all the sensors by a wide margin.
5) Medium-format digital cameras have become affordable as well.
Video at 400k ISO 50fps:
https://youtu.be/1-HfWCVQsIY?t=182
or Sony alpha A7S II - 4K low-light ISO comparison movie
So while there is are small incremental improvements in sensor tech (quantum efficiency, etc), the big development in quality is pure physics. The phones takes as good pictures as compact digitals, because they are now using the same size tech.
Also the picture quality isn't just dependent on the sensor. Better processors can help with autofocus, noise reduction etc, and that obviously helps the end result.
This however will lead to a problem with physics: you can't grow much beyond where it is now, without making the phone much thicker to accomodate a bigger lens, at least not without using exotic thech such as diffractive optics etc. So don't expect the massive gains in image quality over the next 3 years, as in the past 3.
> Better processors can help with autofocus, noise reduction etc, and that obviously helps the end result.
These processors could be put into bigger camera's too I imagine.
As you said, those physics are hard. I'm no expert but I suspect that the physics would even be strictly impossible for a "proper" lens and they only get away with it because they can now apply corrections computationally which would require lot's of glass/space when done to light pre-sensor instead of to data post-sensor.
This, I think, is the actual innovation: when you don't have to care for distortion or uneven brightness because you can fix that digitally, basic sharpness becomes much easier to achieve. A truly raw sensor readout from those modules would probably look like worse than a Holga picture.
I fully agree with your assessment that this is a one-time free lunch. Looking forward to a computational compact that scales the same lens simplifications to a full frame sensor!
I'd also be happy to get a phone with a detatchable camera module you could bring when you need it. So long as it's much smaller and cheaper than a camera, it would still be better than carrying a compact camera and a phone (which was the problem these good cameras solved in the first place).
Nokia 9 is like that.
Admittedly, multiple sensors also create more/better source signal to work with and unlike optical path simplification it's an improvement that can be repeated over and over again (just add more), but currently it's main importance is that it is something for buyers to see, without diving into the subjectivity-infested depths of image quality comparisons. I have no doubts that a bad multi-sensor would sell better than a good single-sensor.
Fuji just announced the GFX100 which is a medium format camera in the same body as a Canon 1DS. And there are rumours they will allow for multi-capture i.e. 400MP image out of the 100MP sensor.
But in the middle between something like that and your phone camera the market is surely going to decline.