Comment by phil-martin

This is so interesting! I would like to see the same experiment, but with these variations:

- All the comments removed (the blog post)

- Half the comments removed

- 10% of the comments replaced with slightly incorrect comments

- 10% of the comments replaced with very incorrect comments

- 10% of the comments replaced with slightly incorrect comments

- 90% of the comments removed, 10% remain but are very incorrect

And variations like that

Test to see where the inflection point is for correctness, token usage and other things