Thanks for catching that! You're exactly right, and I'm not sure how I managed to leave that out...

Your modification is definitely an improvement. I believe the original paper updates the target network after each step. I'll update the code snippets, and re-run some Colab experiments to update the performance metrics.