the kneejerk reaction of "just RLHF it away" to every issue with LLMs isn't great either imo

there is practically no way to tell if it's a homework question or not, given the student is smart enough to at least remove any paraphernalia

maybe, but my latest paper is partly on how openai didn't ever bother to proofread their evaluation code and fucked it up for a lot of languages back when they released Whisper so like

not saying there aren't people working on it, just that they're severely outnumbered by the people throwing sphagetti at the wall and seeing what sticks, at least in the academic circles I've seen

one of the big issues with AI research is the lack of people stopping to evaluate the possibilities and limitations of the new funky model

It's generally true that long-standing experts will be conservative in predictions like these though. True in the case of climate change, scientists often downplay the effects if anything, every climate model update is "whoops it was worse than we thought!"

to the extent with which several SE paradigms are generally unnecessary for research code, its still good to know. judgement call to be made there but

to an extent by the time they're at the PhD stage they *should* have already been exposed to enough code to figure out what good programming looks like. lots of online resources aimed for SE apply pretty well imo but direct mentorship would work too