I can't really comment on the allegations regarding AmadeusY, but I also agree the program appears to be in a rather poor state.
A Beta is normally used for polishing, fixing issues, implementing feedback, and optionally some small new features as long as they don't break compatibility with previous versions.
Anything that does so, and all large features in general, should be implemented during the Alpha stage.
I've written about this on Discord before, but I still wonder what they did with their AI. Assuming they use an NNSVS-like synthesis pipeline, they are going to have three separate AIs (timing, pitch, and final sound), and the program produces errors with all of them.
However, this isn't really typical for an AI that wasn't trained as much, since that gives you (engine) noise, not errors. So I think it's possible they're dealing with an AI that's not well configured in some way.
The process for finding the best configuration is called Hyperparameter tuning, and it needs even more compute power than training. It either takes really long if you do it on a PC (if that's feasible at all), or it's expensive if you rent a server for it. It also means more programming work.
So my theory is that this really is a passion project, they wanted to save time and guessed which configuration might work. (This can often work btw, but it's always a gamble)
But if the issue really is AI configuration, this would be extremely bad for them, since you have to restart AI training from basically zero after a configuration change.