A pro-AI fork would be nice for both "sides"; we could experimentally see whether it helps or hurts. In a few months-years, we should be able to see whether it's faster/buggier/better/worse.