Does anyone know how Norvig thinks about the use of LLMs for programming?

I was curious about this too, so I went looking. I found this video from about a year ago. He seems cautiously optimistic.

The second half is a Q&A and he directly addresses this: in the presence of errors he finds more faults in the way the prompt was worded than in how the LLM answers, and figures that LLMs are better than the alternative approaches to programming if used well.

https://youtube.com/watch?v=ia6aJIplmtc