In the JEP to discuss this, there was agreement that human-readable words were preferable to randomly generated hashes. However, the issue made clear that we need to be very careful about the corpus of words that we use. This is an issue to track potentially choosing a better corpus of words, and adding back in human-readable auto-generated cell IDs.
When https://github.com/jupyter/nbformat/pull/217 is merged, we'll start generating hashes for cell IDs, rather than using combinations of human-readable words. This is because this issue pointed out that some combinations of words were problematic.
In the JEP to discuss this, there was agreement that human-readable words were preferable to randomly generated hashes. However, the issue made clear that we need to be very careful about the corpus of words that we use. This is an issue to track potentially choosing a better corpus of words, and adding back in human-readable auto-generated cell IDs.