Yes, it is interesting to see how such an LLM generates code. However, I found some unmotivated refactoring, as well as scattered changes where, for example, instance variables were replaced by parameters, which is a difficult source of errors to find.
Also, an obvious hallucination that wouldn’t go unnoticed with a decent IDE.
Original:
self._write_file(f'{self.folderPath}/__index.md', text)
LLM:
self._write_file(f'{selfaveringfolderPath}/__index.md', text)
Then the LLM contradicts itself: In the first iteration, it claims that the problem is the scan function and replaces it with a regex function. However, in the follow-up comment, it recommends replacing regex with a scanner. In the last iteration, I then see a scanner, which is a cumbersome variation of my code. I also see a test for Zipfile, which seems completely unnecessary to me.
Something like this shows me that with the LLM we have a counterpart who pretends to have superior knowledge, but in reality understands nothing.
The final version of the code is quite pimped out and a bit opaque and may or may not work for a particular example, but I don’t want to have to look for bugs in it. Instead of fiddling around with the code until it somehow works, I still stick with the good old programming skills.
Whatever the case, can you use it to convert the “Murder on the Orient Express” example satisfactorily?
