As someone on the forefront of ML, you would expect me to be in a position to reap the benefits. It's possible I am incompetent. But I often wonder what we're doing, chasing gradients and batchnorms while training classifiers to generate photos of lemurs wearing suits.
I try not to dwell on it too much, since I truly love the work for its own sake. But one must wonder what the endgame is. The models of consequence are locked up by companies and held behind an API. The rest are nothing more than interesting diversions.
I've been reading some history of math and science, and it seems like many of the big discoveries were made from people pursuing the work for its own sake. Feynman loved physics long before physics became world-changing. But if physics never lead to the creation of the bomb, would it have been so prestigious?
We seem to be lauding ML with the same accolades as physics during the postwar period. And I can't help but wonder when it will wear off.
ML will be a fine tool for massive corporations, though, for endless reasons. But I was hoping for a more personal impact with the work. Something like, being able to enable a blind person to use a computer in a new way, or... something more than memes and amusement.
Perhaps doing the work for its own sake is enough.