I'd consider checking out some of the YouTubers like Alex Ziskind with his benchmarking channel and so on. The wild card is the amount of local AI and the scale of models you might wish to run locally. Living with networked AI for a while drastically lowers the stakes. Ziskind has some great stuff on benchmarking various models to give you the basic nuts and bolts of how fast hardware is with what AI model which is useful for knowing what can be run locally, while Simon Pittman (Youtube channels Systems Made Better and Better Creating) has some very good, clearly explained tips on integrating networked AI on fairly basic hardware.
As for the display size, go to the store and pick up the computers if you can, if you're near an apple store in particular. See what you're willing to put up with. I find the true two-page displays extremely useful, as well as cooling power, and in a Mac laptop that's mainly the 16, maybe the 15 Air at a pinch. But the smaller models -- even the Neo -- are also proficient at running at least one external display (a Max can run four high-res displays) so if you don't need two pages when away from the desktop, that's a factor. Another factor is if you have a desktop computer or not. My laptop is my primary but for those with a primary desktop or doing a lot of always-on computing with agents or something like that, it can make sense to get separates.
Also find out what your program/school/college/university recommends in computing. Medical schools, for example, were really strongly pushing their students to get smartphones 25 years ago. A CS program with very strong or very weak included online resources might have some interesting variations on standard requirements. 16GB to 24GB RAM on a base M5 or M5 Pro can get you a surprisingly long way if you just used networked tools. The AI market is also rapidly evolving in unpredictable ways; nobody was calling out this year's push for agents last summer. Local AI is another wildcard; because even bloatware like Adobe Photoshop runs fairly happily on a 16 to 32GB M1, but local LLM and suddenly 64GB to 128GB are needed for many models (and some need much more even than that), and the Max means more memory bandwidth.
If you go Max I'd tend to get a 16 for the better cooling. Also sort out what tasks are bandwidth limited, compute limited and so on, that may affect decisions. Bandwidth limited and it will take advantage of fast memory but may never tax the processor and then the 14 Max makes sense. Compute limited and it absolutely will get those fans humming. But a 15 Air is also a possibility with sufficient RAM and storage.
