I'm unnerved by common claims of "with our super duper library X, doing Y is just 5 lines of code: ...". Ok you hid a lot of code under defs and reduced flexibility. I'd rather see it be 50 lines of code, with nice modular building blocks where various reconfigurations are clear.
— Andrej Karpathy (@karpathy) February 22, 2020
People who complain that information security is terrible and getting worse must have short memories. Only 20 years ago, malware was unstoppable. It seemed a Very Bad Idea for anyone who wasn't a security expert to use a computing device for anything important or sensitive.
— Arvind Narayanan (@random_walker) February 22, 2020
On (1), eager mode was fine until 2 years ago -- with increase in register file, flops, interconnect, a good compiler will become essential
— Soumith Chintala (@soumithchintala) February 21, 2020
On (2), {a general purpose API} and {specialized packages for a subset of algorithms} will always cycle
Tools have to evolve!
(2/2)
Similarly, people greatly overestimate the benefits of brevity in writing, especially on the web. When an essay is riveting, people want more, not less. Tolkien's criticism of his own "The Lord of the Rings" was apt: "the book is too short". https://t.co/uKvAJbLiOI
— michael_nielsen (@michael_nielsen) February 20, 2020
Enjoyed this piece by @a16z on the challenges of applied AI.
— Emmanuel Ameisen (@mlpowered) February 18, 2020
They call out the difficulty of choosing the right model for a product.
I agree and cover ways to do so in the first chapter of BMLPA.
It is available for free at https://t.co/sZFAcgrQVFhttps://t.co/WK0WwuJVmj
https://t.co/oS61GmG4W8
— Brandon Rohrer (@_brohrer_) February 18, 2020
Excellent read if you are a data engineering manager or thinking about it.
From @tayloramurphy
h/t @vboykis
While it's great to see new AI algorithms using human intelligence as the baseline, I just want to bring back this Edsger W. Dijkstra's quote:
— Chip Huyen (@chipro) February 17, 2020
"The question of whether a computer can think is no more interesting than the question of whether a submarine can swim."
When time is limited, preparing for rigorous coding/puzzle interviews at large firms can take many months (especially for people like me who were not that good technically).
— hardmaru (@hardmaru) February 17, 2020
The same amount of time might be better spent working on projects that define the essence of who you are. https://t.co/vJuUe4Lql4
The Dystopian World of Software Interviews
— hardmaru (@hardmaru) February 15, 2020
“Complaining about it wont do any good—Sooner or later I’m going to have to get another job… For now I am going to choose the only option that leaves me with my sanity:
I’m going to exit the loop for now.”https://t.co/fyVCbozGnV
this is important.
— (((ل()(ل() 'yoav)))) (@yoavgo) February 13, 2020
convincingly pointing to a hard and important problem is far more important than solving an easy one. https://t.co/eobhC4U1HB
A lot of "X is making people stupid" moral panics boil down to "people were stupid all along, and X is making that painfully obvious"
— François Chollet (@fchollet) February 8, 2020
While I cannot advocate for or against anyone attempting to make a living from ML with non-traditional background (it's their decision to make) there are two common themes that come up over and over again in success stories:
— Radek Osmulski (@radekosmulski) February 6, 2020
✅persistence
✅power of personal connection
(a 🧵) https://t.co/Ws2PWuYfKf
