952 karma · joined June 8, 2019
We have a big problem with that where I'm from. We get a lot of brain drain going over to California (ok fine, everyone does), not necessarily because of the opportunities that are undeniably present there but because early-career tech people here are inherently less willing to accept the proposition that it's possible to make the work cool where you are; not everyone will work for Google Brain once in the valley, anyway.
I struggled greatly with this and had a big misadventure going to the US myself but the perspective it gave me on the need to lean in(wards) was worth it. Hope it is for someone else too.
Not trying to knock this by any means, I hope it works wonderfully!
I'm reminded of that joke:
Business logic (n): your stuff, which is shit, as opposed to my code, which is beautiful
I say this as a machine learning engineer
Long story short, if you want to have a cache, have a cache. If you want to make something parallel, make it parallel. This stuff just gets in the way. The only real exceptions I make for decorators by and large are for pydantic things and similar. It's very noticeable that if you're doing stuff in pytorch lightning you don't tend to fiddle with decorators much at all but in tensorflow it's common. Cute code isn't good code, end of story
Julia has always had a reputation in my mind at least of being "by academics, for academics" and there's unfortunately a dark side to that in terms of reliability and maintainability. The concept and goals are great, which is annoying. If this language had stayed focussed on the basics, it would be extremely handy for someone like me who trains and deploys models in an edge computing environment. No way I'm doing that with stuff like this going on.
As it stands, it seems like your algo will always use the same first guess, have you thought about making that configurable and seeing what happens with e.g. the number of words where it gets stuck and runs out of time?
Who's next? Get out of that as fast as possible
Enough people think like that, hey presto
How does Tiny manage to stay non-interventionist as they claim?
It's interesting but I think there has to be more than what's pointed out here going on
Disintermediation of data pipeline creation is definitely nothing new at this point and the technologies aren't that novel at this point either. I'd be surprised that this is on the front page, but it takes time for the lessons in this article to be learnt by a large enough amount of people that it becomes humdrum.
Above all, it reminds me of a consultant friend telling me he had two clients who built feature stores - one with an open-ended goal of enabling people and one because they had some specific things they wanted to achieve. The outcomes they got were as dissimilar as their motives!
Part of solving the problem is legislation, part of it is cultural. If you don't have both, then sure, it's too hard.