There's clearly a demand for having an AI go through every public PR in terms of code quality and security. But something generic like ChatGPT-4 doesn't seem to be capable of catching this from just looking at the diffs.