Since genetic engineering is necessary to first put light sensitive ion channels in the brain, I don't see how this is "way better" if you're targeting applications in humans like neuralink.
Comparing few data points to none just doesn't allow for any intuition about the safety of either compared to the other.