I felt that when I gave my GPT-3 agent tools for both manipulating & reading lists of data, and to my surprise the agent decided on its own to read a list after

I felt that when I gave my GPT-3 agent tools for both manipulating & reading lists of data, and to my surprise the agent decided on its own to read a list after modifying it, in order to do validation. So in other words, this mf (GPT-3) didn't trust its tools' code and apparently felt it had to check my work. But, similar to what this blog post shows, it turned out I actually did have a bug, and the agent didn't care! It just "thought" to itself that it verified the correctness.