RT @kearneymw: This tweet is gonna seem genius one day when the innies in Severance are revealed to be labelers.

@threadreaderapp unroll.

What do you want to know?.

RT @SolomonMg: 🧵Do not accept the narrative that Trump won b/c racism & sexism:. 1. Doesn’t line up with the evidence.2. May embolden white….

I’m done with Twitter. You can find me at

It’s weird that a desire for cheaper groceries would result in support for….- 60% tariffs on Chinese goods.- 20% tariffs on all other imports.- Mass deportations (labor shortages).all of which are likely to *increase* inflation.

First the betting markets hard correct. Now Flacco disappoints. I should probably be saying this stuff on bsky so people know how good I am at pointing out obvious things.

Their fast turn around time and efficiency might be explained by their weights, which not for nothing did quite well in 2020. They might be making some smart assumptions that other pollsters aren’t, and the consistency is explained by those weights and the responses they get.

There are three ways to serve results from a “model.” The first way is to pre-calculate all answers and perform table lookups. The second way is to perform the inference in real time. The third way is to use cached results from a table and real time inference for the rest.

Happy Halloween!

When we see evidence of herding in the polls it’s tempting to turn to other things like betting markets, Joe Flacco, or Rob Johnson, because they seem to offer clarity. But any clarity they provide is fleeting. And we’re only left with regret.

But when these pressures invariably catch up to these grass-is-greener QBs, teams are often left worse off as they’re suddenly relying on an option they originally knew to be less effective. And this is not unlike our current situation.

This is also why so-so NFL QBs like Joe Flacco, Andy Dalton, and Gardner Minshew often play well in their first games back as starters. With less pressure and no recent track record to constrain them, they play looser and more freely.

This means seasoned, objective pollsters can make a series of smart, reasonable decisions—each defensible on its own—that are subtly shaped by the pressure to be ‘right’ and awareness of other polls.

Quick note on election poll herding: Strong evidence of herding doesn’t imply pollsters are manipulating numbers. Often, subtle, subconscious shifts happen under pressure and from heightened awareness of other polls.

So cute tho.

The baby hijacked my Warby Parker virtual try-on session.

What would you like to understand better about machine learning?.

Data science manager flex: converting a nontrivial amount of polars code into pyspark between meetings.

All data are unstructured if you zoom out enough.

In Rock-Paper-Scissors, do you focus more on your move or your opponent’s?.