One of my all time favorite probability facts.

A loose intuition is that you can't expect that new information (Y) will move your belief about something (X) in a predictable direction, because if so you should change your belief before getting the new information.

BEFORE
- Standard bar chart
- No clue what the message is

AFTER
- Bar chart with nuanced color use
- Informs your reader about key insights & actions

The latter is actually pretty easy to pull off.

At the end of this step-by-step guide, you can do that too. #rstats #dataviz

Neural networks often pack many unrelated concepts into a single neuron – a puzzling phenomenon known as 'polysemanticity' which makes interpretability much more challenging. In our latest work, we build toy models where the origins of polysemanticity can be fully understood.

This graphic about where people go after college is ENLIGHTENING and probably something my dad would have cut out from the newspaper if it was in there in 2002. Andrew Van Dam washingtonpost.com/business/2022/…

My colleague Joe Blitzstein's Stat110 course 'Introduction to Probability' has the highest enrollment of any FAS course at Harvard, 783 (passing habitually top enrollment courses Econ10 and CS50), in Fall 2022. Congrats Joe on making history!

By the time you’re n=30, you should be approximately normally distributed.

This seems right: people resist 'taking COVID seriously' because they fear they'll be asked to make major lifestyle sacrifices. A reasonable fear given the past 3 years. So maybe take-COVID-seriously advocates need to say explicitly they're NOT asking for big lifestyle changes.

I can't tell if this is a bug or a feature in Discord:

I sent a regex string as a joke to a friend, then it actually edited my previous message with the regex!

Me: You are such a pro!
Them: No, I'm not
Me: s/pro/liar

This made my original message change! SO cool either way!

I printed out my data and hung it on the wall. my wife said it was missing something so I put a decorative border around it.

guess you could say I added a dataframe.

Some folks in Silicon Valley have been overly indoctrinated in a gospel of testing and letting data make all decisions.

While that’s a fine ideal, 90% of the time startups don’t have enough time or data to test.

It’s OK to simply talk to your customers and use instinct.

As you become an adult, you realize that things around you weren't just always there; people made them happen. But only recently have I started to internalize how much tenacity *everything* requires. That hotel, that park, that railway. The world is a museum of passion projects.

What are the best ML *books* to come out in the last three years?

They can be code-focused, theory-focused, whatever, but they have to be books.

You guys... I'm genuinely embarrassed that I ever believed that 'Target can predict you are pregnant' story. This was clearly someone bamboozling the company like 'look at how great my fancy model is!!' nytimes.com/2012/02/19/mag…

Many people implicitly assume that if a number shrinks by x%, it has to grow x% to get back to its old value. That's not far off for small x. A number that shrinks by 10% only has to grow about 11.1%. But a number that shrinks by 75% has to grow 300%.

A workshop student asked me for places to go to see collections of great data viz (he said he goes to Reddit's dataisbeautiful 🫤). I asked my LinkedIn followers for their suggestions and this is a thread of their go-to viz inspo spots.

interviewer: can u explain the gap in your resume

me: i use LaTeX

Sometime around 2015 there was a mysterious vibe shift in web design.

Before the vibe shift, all the cool websites looked like this:

After spending many years on this planet working with numbers, today is the time I had reason to find out the quantile term for splitting distributions into thirds.

'A Tertile divides an ordered distribution of values into three parts'.

This is cool: when AI is better than humans, it makes humans better. In 2016, the Go world was shocked when AI beat the best human player. Since then, by playing against AI, professional players have gotten unprecedentedly better at the world's oldest game hyokang.com/assets/pdf/CKK…

In regression, there are several things that students are eternally concerned about but are actually Just Fine:

1. Your coefficients don't need to be significant
2. Your R^2 doesn't need to be huge
3. Your predictors can be correlated
4. Your variables don't need to be normal