AI and machine learning
This interview is interesting enough, but it's really good as a reminder of how early we still are in trying to educate a generation of data scientists and AI experts -- and how we still haven't figured out how the balance between industry and academia should work.
But not yet commercially. That will happen soon enough, though, and perhaps then we'll get a sense of what's actually possible with embedded and even offline AI models.
Just what it sounds like, from Google.
You control your phone's screen by moving your head. Could be really useful, and possibly annoying depending on how it's implemented.
The idea seems sound -- human children are curious and experiment in their worlds, so build AIs that can do the same. The big question I always have around these efforts relates to ambition. Children have innate reasons to explore, going as deep as survival, but it seems difficult to replicate this type of thing in AI.
A really interesting story about how algorithms are helping build a market for kidney transplants.
This is presented as building AI systems that can discern hypotheses and other higher-level tasks beyond keyword or phrasal recognition, but that seems like a challenge. I'm reminded of legal services like Lexis and Westlaw, where the experts summarizing cases and curating related authority is really the value. Depending on the situation, you still can't trust a machine to catch or understand everything.
Breaking down problems into smaller, explainable tasks seems like a reasonable approach to reasoning. Although "attention mask" is not an appealing term.
So it's clear what they're talking about: "Researchers at Carnegie Mellon University have devised a way to automatically transform the content of one video into the style of another, making it possible to transfer the facial expressions of comedian John Oliver to those of a cartoon character, or to make a daffodil bloom in much the same way as a hibiscus would."