Earlier quoted context omitted.
Few people are "blown away" by the absolute skill level of ChatGPT. It's more that there exists an AI that can do this at all . This just wasn't a thing three years ago, and the first truly capable version (GPT 4) only become publicly accessible this year . Some things I've seen GPT 4 do is "as good as human", often better than average. For example, spotting bugs or reverse-engineering code and then writing doc comme…
You want to trust that thing to review your code? It will often give you eloquent explanations, trying to convince you that code has a certain bug, while completely missing the actual bug. You'll be gaslit into chasing down dead ends, distracted from actual issues. Example: "In JS, create a Promise that is resolved with some emitted value from an EventEmitter, or rejected if an 'error' event is emitted first. Write t…
ChatGPT doesn't need to find 100% of bugs in 100% of code out there.
Think of it as a "better linter". It's cheap, and finding just 50% of bugs would be an enormous step up for quality in most code bases.