I am still surprised by the large PIC fan-base, and the fact that xc16 and xc32 are based on gcc that there are no community repository with a pre-compiled gcc without restriction and with proprietary part removed (even if it include some optimization layer. GCC already has more than ok level of optimization for pic32 if its not crippled AKA allow optimization).
By the way, nowadays, if you want to start out pic16 bit, you could also jump straight to 32 bit. PIC24/dsPIC are complex enough at peripheral level that it is not harder than going pic32 or some entry level/easy to use arm like the entry level lpc from NXP. It is more a matter of taste from now on, and a balance between feature, speed, predictability (interrupt latency, branch, pipeline...), and stability/maturity.
If we compare PIC24/dsPIC to PIC32 (since you know better microchip product):
PIC24 is better for hard real time, since it has a less complex pipeline, more predictable (every if statement will always take the same number of cycle, on pic32, because of branche optimization, wait-state, and other such optimization mechanism, it may be sometime faster, sometime slower). PIC24/dsPIC is a more mature product, less bug than PIC32. PIC24/dsPIC has better math performance in tight loop (for doing FFT, and other dsp related work). On the long run, PIC32 has better performance in big loop/multimedia (because of all its peripheral for multimedia, like lcd controller, 32 bit math for doing audio decoding). PIC24/dsPIC has faster GPIO, more peripheral related to motor control, reading encoder... PIC32 has more memory. But in the end, they both have the same complexity. ARM is like 10-25% harder compared that (as long as you don't take stm32f4/f7 or other big number crushing arm with advanced peripheral).
Oh and another factor important : your style of literature. A factor that few people take into account but can make a huge difference. Some people understand some author better. For example, I can't understand many datasheet from microchip (how the data is formatted, explained, fragmented between multiple document, from my opinion, data in the wrong place). But I can understand datasheet from texas instrument (msp430) and atmel(AVR & sam) without problem. I know many other people that just can't stand atmel datasheet and feel right at home with microchip datasheet. But it also depend on who written the datasheet. Sometime, the same author written the datasheet, 20 application note, and 2 user guide, and many peripheral manual, except the one peripheral you need, that was written by someone else who you have difficult understanding what he say. Before choosing your target for your next few month of playing, explore different micro-controller, datasheet, and try to find your sweat spot of understanding in those datasheet.
Oh and by the way, in no way I say you SHOULD go 32 bit, I just say it is an option you could consider. Unlike many other people, I don't say that 32 bit is the solution to everything, I see its advantage and disadvantage. But I agree 8 bit is really outdated and people should stop using it, except if they got an overstock of some part. Especially the 18 bit pic family. Nowadays, 16 bit are so much more better, cheaper, more featured, more efficient (less wasted power). Except from utterly simple task like electric toothbrush, where a 4 bit could suffice, people should move on. 16 bit is not dead, and are more than enough for any task, including quadcopter.