It is 2022 all over again!.
Back then, developers on this site (including me) said it was literally impossible for AI models to write working code.
Now, people know writing code is solved*, so they're moving to the next frontier: reviewing code is impossible!
It's insane you and others say this with a straight face. You realize code reviews are the hardest part to do well (not bs reviews that only look at surface level charges and auto click merge). Non-trivial/large reviews are the hardest part of software development that require actual intelligence and a huge amount of context as a "review" requires the ability to think critically... A good reviewer also has the ability to consider if a pr could just flat out be implemented better in a different or simpler way. Current agents are terrrrrible this especially in large codebases. It's is an impossible task unless you have full agi or a something capable of novel thoughts. LLMs flat out today do not have enough context to be able to do complex code reviews well. Making a good llm based reviewer will never work for literally the same reasons LeCun says LLMs can't further scale.View quoted post on X
