Over the last month, in my spare time, I’ve been experimenting with two separate ideas: some recurrent-network maths as a possible GRU/LSTM substitute, and a separate transformerless AI project. The transformerless project eventually became too slow to…