I would love to know more about this. Even with other SOTA models doing a "review". It won't find anything and then Copilot Review will find 3 other things. It seems weird that they swing from yep we fixed everything to completely missing two of the same columns in an SQL query.
We've replaced it with a bot that does the same thing, except we fully control it now. So we can route crucial parts of the codebase to SOTA models, tag security if needed, etc.