What the heck?

How come AlphaBet can fold proteins but cannot predict my next video?

You’ve seen it everywhere: Alphabet's Deepmind’s program AlphaFold made a breakthrough in a very hard problem of biology: predicting the 3D structures of proteins.

Essentially, AlphaFold is state-of-the-art machine learning, developed by a holding of companies like Google and YouTube that also use machine learning to assist searches and video suggestions. But something is broken in the latter…

Youtube fails miserably at linking videos of a series when they are not in a playlist! For example, I’m watching “Series X episode #1”, and when this video finishes the system auto-plays “Series X episode #4”, or #3, or #6, and seldom #2. And this happens systematically with every series!

So, AlphaBet can fold proteins but it cannot understand video titles well enough to grasp that there is a sequence to follow?