algorithms and data structure questions in interviews for most software engineering jobs are a sign of laziness of the interviewer preparing for interview. Turn away from such place for two reasons: https://t.co/awGSTT726k
— Delip Rao (@deliprao) July 26, 2019
Research debt comes in many forms: poor exposition, undigested ideas, bad abstractions & bad notation, noise.
— Rachel Thomas (@math_rachel) July 25, 2019
The insidious thing about research debt is that it’s considered normal. People don't realize that things could be different.https://t.co/ge1vcnFd05 @ch402 @shancarter pic.twitter.com/x46OBedq4j
Calling data science "AI" is like calling a crane or drill a "construction robot"
— David Robinson (@drob) July 23, 2019
Tools might make it possible for us to do remarkable, previously-impossible things, but they're still tools
To solve a hard problem, you need to be pretty much obsessed with the process of solving it, not merely enamored with the idea of having solved it.
— François Chollet (@fchollet) July 22, 2019
There are many people in the latter category, rather few in the former
The cognitive fallacy that tech folks seem most likely to fall for is believing a broad & complex problem can be solved with a single, simple universal solution (backed with a couple of cherry-picked examples)
— François Chollet (@fchollet) July 22, 2019
Contrary to popular belief, training a gigantic model on a humongous dataset of human text will not lead to AGI. 🙈🧠
— hardmaru (@hardmaru) July 22, 2019
Probing Neural Network Comprehension of Natural Language Arguments: https://t.co/3AtozIV03O https://t.co/Y5zg0Dfgke
My advice to graduating CS students:
— Arvind Narayanan (@random_walker) July 21, 2019
Any individual employee's efforts to act ethically will be steamrolled by the effects of the company's financial incentives.
But collective action by employees can make companies change direction. We've seen it in many tech co's recently.
10x engineers are really 1x engineers working together in an environment that simply lets them do their jobs https://t.co/xRGEuCobYk
— hardmaru (@hardmaru) July 21, 2019
Every supervised model, in general, “overfits to spurious statistics” (note overloading of “overfitting”). The folly is in thinking that we can just do supervised curve fitting and somehow a causal concept (e.g. spuriousness) will magically fall out. #bookofjudea @yudapearl https://t.co/1FjjvKx0ND
— Zachary Lipton (@zacharylipton) July 21, 2019
The older I get, the more I appreciate timing as a force multiplier.
— Chris Albon (@chrisalbon) July 21, 2019
The right argument, made at the right time, can be 10x more powerful than the same argument made at a different time.
As a manager you have a major effect on people’s overall mental health. A poorly phrased sentence (or a joke about firing someone) can really wreck someone for weeks. Too few managers are conscious about this or care. https://t.co/fk3Lt5il6l
— Chris Albon (@chrisalbon) July 20, 2019
This touches on what makes me squirm every time I teach the ML paradigms as "supervised, unsupervised & RL". While the framing is sometimes useful, it's a gross category error. SL can be cast as special case of RL, and auto-regressive learning is both supervised and unsupervised. https://t.co/oyjZZRsZO1
— Zachary Lipton (@zacharylipton) July 16, 2019
