sundalia

7 ms·

1.
by sundalia 2y ago
Application-specific metrics are the way to go. For ML training this is one example: https://cloud.google.com/blog/products/ai-machine-learning/g...
2.
by sundalia 2y ago
AFAIK all efforts in that direction were way too costly a few years back and degraded models considerably. Spotify, for the longest time, only trained on an equivalent of manually curated playlists by experts and users to understand similar
3.
by sundalia 2y ago
Yup, we can close the thread and ack that GitHub does not care.
4.
by sundalia 2y ago
There's good points the author makes about problems in interviews + some economical facts but pieces like: > Sorry, you’re not qualified to be a professional software developer because you wore the wrong color shirt to the interview
5.
by sundalia 2y ago
(?)
6.
by sundalia 2y ago
Then don't make claims you can't justify :)
7.
by sundalia 2y ago
We're talking about a company's actions, why does vote matter?
8.
by sundalia 2y ago
Would appreciate data.
9.
by sundalia 2y ago
Many Googlers are not comfortable with that. Lots of them are immigrants who understandably don't want to be associated with US-endorsed genocide.
10.
by sundalia 2y ago
>> It’s not participating. I wonder why
11.
by sundalia 2y ago
Unique? You mean tweets? Yeah sure It's 6B down the drain. Saying grok 1.5 is competitive is a joke, if it was any good it would be ranked well in chatbot arena ( https://chat.lmsys.org/ ). Elon is a master in hyping un
12.
by sundalia 2y ago
This is very interesting. A few questions: - Why do you think similar approaches never landed on jax? My guess is this is not that useful for the current optimizations in fashion (transformers) - How would you convince jax to incorporate th
13.
by sundalia 2y ago
How is it serious if money is the motor of freedom of speech? The suing culture in the US ensures freedom of speech up until you bother someone with money.
14.
by sundalia 2y ago
+1, this is just the commenter saying what they want without an actual court case
15.
by sundalia 2y ago
Quite frankly, I see a lot of text in this post and no numbers. For something to be production-ready I'd expect you to at least cover major things like "latency to serve x in Elixir instead of lang y is k% better" or "EM
16.
by sundalia 2y ago
Jeremy Howard comes to mind. The sheer good that fast.ai brought as free quality educational material feels enough for "good in general" :)
17.
by sundalia 2y ago
>Some friends told me they find this blog post mildly sociopathic These are likely tech friends I suppose? I would say non-tech people would drop the mildly. This is leaking the productivity frenzy mindset through the cracks, but appreci
18.
by sundalia 2y ago
Absolutely would not work in the US (as, unfortunately, most public services). There's already a huge lobby from TurboTax and other players. All countries that got to this level of automation already had publicly developed, free softwa
19.
by sundalia 2y ago
Not everything has to be clear and to the point. The sentence's murkiness adds emotion to the text. I'm not a native speaker (just like the writer isn't) and maybe that's why we're not so bothered. But this comment
20.
by sundalia 3y ago
The intro query "youtube downloader" already showed me relevant results (some website where you paste an URL and bam download). I think there's a big tech bias in the whole post (how relevant is a mastodon poll, for real). No
21.
by sundalia 3y ago
fast.ai (the first iteration) was what got me into ML, when I was a sophomore undergrad. It played a big role in my career choice and progression! To this day it still is my first recommendation for those learning. Congrats on the launch,
22.
by sundalia 3y ago
They picked Portuguese from Europe but omitted Brazilian Portuguese (largely more spoken, like 90-10 split). It's not ananas there, it's abacaxi.
23.
by sundalia 3y ago
Empathize with parts of the post, specially on the turning 30 part. However, this smells a lot of junior level and poor managers. The git rant, the 3 pillars of being a good swe, the working code trumps good code ... All thoughts feel in th