Loading…
32
Am I the only one who remembers when a neural net couldn't even get a stop sign right?
I was digging through some old bookmarks last week and found a YouTube demo from 2014 where a computer vision model mislabeled a stop sign as a speed limit sign because of a little sticker on it. Back then that felt like a magic trick, and people in the comments were losing their minds over it. My nephew is 19 now and he shrugged when I showed him, because his phone does that in its sleep. What got me was the number of images those old models trained on, something like a million, and now a kid on a laptop in a dorm in Columbus can fine tune a model on a few thousand pics and beat it. It hit me that the gap between wow and boring keeps getting shorter, and I am not sure how I feel about that. I still have a soft spot for those clunky early demos, mistakes and all. Does anyone else keep old papers or videos around just to remember how weird the starting line looked?
0 comments
Log in to join the discussion
Log In0 Comments
No comments yet
Be the first to share your thoughts on this discussion.