GPT did an interesting thing for me just now. I am rejiggering parts of my book. Having done that, I discovered that I developed two versions of what is essentially the same argument in two different places. But, now that the macrostructure of the text is more coherent, those two versions find themselves in the same place. They are similar but have slightly different emphases and make slightly different, albeit complementary, points.

So I asked GPT to do something I thought it was capable of doing: splicing the two versions together so that corresponding parts of the text would be placed together. I could then decide which I wanted to keep, where. I asked it to preserve my text perfectly and [[put double brackets]] around the bits IT would cut (showing which version I might keep.

Turns out, this task was pretty much beyond ChatGPT 4o, at least after 4-5 iterations of explaining myself etc. BUT ChatGPT o3-mini-high was able to do it pretty well on the third try.