It works for local+streaming users, but not local-only, because the algorithms require machine learning models that use more music and plays than you have in a user-centric library.
If they were trimmed to use only your limited local library, they would fail to give any reasonable results. The old algorithm is good for local libraries, but it’s pretty bad for discovery of new material.
Daily mixes are driven towards discovery, not just shuffling – no matter how smart that shuffle is.