Exciting to see work on an nn that would be capable of continuous learning by design. The possibilities seem endless - models that can be passed around and built up in a variety of contexts, not to mention the ability to train incrementally rather than spending a few hundred thousand renting A100s. You could leverage the vast unused compute on personal workstations around the world.
1 comment
[ 2.8 ms ] story [ 16.1 ms ] thread