I work on ML at FAANG I use LLMs a couple of times a week (daily if you count line auto complete) It's not that LLMs are bad, but they are just one topic in th

I work on ML at FAANG I use LLMs a couple of times a week (daily if you count line auto complete) It's not that LLMs are bad, but they are just one topic in the vast world of different things. There have been so many amazing insights on HN from users with such interesting backgrounds (nuclear engineering, bakers, musicians, etc). And it just feels like that's all collapsed into "have you tested model-version-quant-subversion-domain? If you haven't you can't say anything"