The crux of the social issue that has been brought about by our new LLM program is that current and future generations of learners have regressed to a more primitive form of learning, where they see a final result produced by an opaque process, and have to reason about whether it makes sense. Throughout human history, the pedagogical tradition that has be proven to work is to learn reasoning in a bottom-up manner; i.e. learn to reason about correctness starting with reasoning about simple things, and learning that complicated things can only be correct when it is a correct composition of correct simple things. Of course, there's also pressure to produce work quickly, and people who succumb to this pressure to produce sub-par work have been doing it before the new program, and the new program simply makes it easier to continue doing what they were doing. The issue at hand applies to people who genuinely want to do good work, but are not able to reason about why a short-cut that produces something "plausible" and passes mechanical checks is nonsensical. Take my summer intern for example; they were not able to understand why the code they generated was nonsensical, as it "made sense" to them, and passed all mechanical checks. Only someone with a lot of hard-earned experience (me, in this case) will be able to point that out, and a green person's attempts are fundamentally futile — it's a bit like trying to learn from a huge example on StackOverflow. We get a lot of such PRs in LLVM, but it's not easy to explain to each contributor that they are fundamentally doing things the wrong way. As reviewers, we have very limited bandwidth, and decide to ignore it after a quick 3-second glance.