Took a lot of time. Right now it is now or in August.
#HAL https://t.co/4DwnS5iLcW

I've spoken to 1,500+ people about remote work in the last 9 months

A few predictions of what is likely to emerge before 2030

[ a thread ] 💻🏠🌍

🚜 Rural Living: World-class people will move to smaller cities, have a lower cost of living & higher quality of life

These regions must innovate quickly to attract that wealth. Better schools, faster internet connections are a must

⏰Asynchronous Work: Offices are instantaneous gratification distraction factories where synchronous work makes it impossible to get stuff done

Tools that enable asynchronous work are the most important thing globally remote teams need. A lot of startups will try to tackle this

⚽️Hobbie Renaissance: Remote working will lead to a rise in people participating in hobbies and activities which link them to people in their local community

This will lead to deeper, more meaningful relationships which overcome societal issues of loneliness and issolation

🌍Diversity & Inclusion: The most diverse and inclusive teams in history will emerge rapidly

Companies who embrace it have a first-mover advantage to attract great talent globally. Companies who don't will lose their best people to their biggest competitors

November is here, and that means a massive shift is coming. And by "massive" I am of course referring to the redefinition of the kilogram unit of mass that the world has been building up to for more than 100 years. Let me explain:

1/ I've had an unhealthy fascination with metrology (the study of measurement) ever since my 2nd year as a physics major when I took a class devoted to duplicating historic physics experiments, so please indulge me for going into heavy detail (get it?) about the kilogram.

2/ So what actually *defines* a unit of measurement? If you're American, you probably know a mile is 5280 feet and a foot is 12 inches and an inch is 2.54 centimeters etc. But where does this chain of definitions end? Is it turtles all the way down?

3/ It's actually not! For all units (even the imperial units used in America) the answers all end with the Système International (SI) unit definitions established and maintained for over 100 years by the Bureau International des Poids et Mesures (BIMP) in France.

4/ At the base of this tower are the SI base units. Just 7 SI base units define every other unit in existence. They are:

Kilogram, kg (mass)
Meter, m (distance)
Second, s (time)
Kelvin, K (temp)
Ampere, A (electric current)
Candela, cd (luminous intensity)
Mole, mol (quantity)

Important paper from Google on large batch optimization. They do impressively careful experiments measuring # iterations needed to achieve target validation error at various batch sizes. The main "surprise" is the lack of surprises. [thread]

https://t.co/7QIx5CFdfJ

The paper is a good example of lots of elements of good experimental design. They validate their metric by showing lots of variants give consistent results. They tune hyperparamters separately for each condition, check that optimum isn't at the endpoints, and measure sensitivity.

They have separate experiments where the hold fixed # iterations and # epochs, which (as they explain) measure very different things. They avoid confounds, such as batch norm's artificial dependence between batch size and regularization strength.

When the experiments are done carefully enough, the results are remarkably consistent between different datasets and architectures. Qualitatively, MNIST behaves just like ImageNet.

Importantly, they don't find any evidence for a "sharp/flat optima" effect whereby better optimization leads to worse final results. They have a good discussion of experimental artifacts/confounds in past papers where such effects were reported.