Ian Barber examined a research paper on test-time training and argued the label covers two different things. The paper is titled "Test-Time Training with KV Binding Is Secretly Linear Attention". Its authors say such designs can be rewritten as a learned form of linear attention. That contradicts the usual view that these layers memorise key-value pairs while running. Barber distinguished slow weights, which are updated by ordinary training, from fast weights, produced during a single pass. The paper trains fast weights, using a matrix that does not grow with sequence length. Barber noted the fast weights reset for each new input sequence. So the approach does not deliver continual learning, in his reading.