@karpathy

high engagement X post #Neural architectures for time series

Explained how transformer-based models outperform LSTMs for long-horizon time series forecasting by capturing global dependencies in sequential data.