LEARNING AT TEST TIME

Summary from the TLDR AI newsletter

A new RNN variant that outperforms Mamba on a number of tasks. Importantly, the update function is itself an ML model - which enables long contexts and in-context learning.