Reddit discusses RNNs vs Transformers vs SSMs for continual learning

Post argues the key difference is where memory lives: RNNs use a recurrent state, Transformers use a growing KV cache, and SSMs use the model network. Suggests each has trade-offs for continual learning.