"While I don't doubt you've given talks about applying ML, I do have a little skepticism that the talks went any deeper than just application-related topics."
Let me correct that misconception; every aspect of the algorithm was custom, from tokenization through classification through smoothing and postprocessing. It was a more simple model and process, but it was not exclusively application related. Far simpler than many of the deeper models I consume as a black box, for sure, but I'd defend my competencies in that I'm not exclusively a plug-and-play data engineer :)
Let me also emphasize that I intentionally called out "driving" as opposed to "fixing". In fact I can't think of any car-fixing research going on just about anywhere, although I'm sure there is. My core point was that _using_ the tooling is key, and has been democratized in both driving and AI (the application layer aspect as you put it), the research point was merely an add-on that I've also seen a democratization of the knowledge needed to do even the basic level of novel algorithm development, even if I'm writing deep network papers or something of that ilk; I just find that to be the exception rather than the rule (per my "hardcore research is hard for the non-deeply-initiated once low hanging fruit has been pruned" comment)
Let me correct that misconception; every aspect of the algorithm was custom, from tokenization through classification through smoothing and postprocessing. It was a more simple model and process, but it was not exclusively application related. Far simpler than many of the deeper models I consume as a black box, for sure, but I'd defend my competencies in that I'm not exclusively a plug-and-play data engineer :)
Let me also emphasize that I intentionally called out "driving" as opposed to "fixing". In fact I can't think of any car-fixing research going on just about anywhere, although I'm sure there is. My core point was that _using_ the tooling is key, and has been democratized in both driving and AI (the application layer aspect as you put it), the research point was merely an add-on that I've also seen a democratization of the knowledge needed to do even the basic level of novel algorithm development, even if I'm writing deep network papers or something of that ilk; I just find that to be the exception rather than the rule (per my "hardcore research is hard for the non-deeply-initiated once low hanging fruit has been pruned" comment)