Happy May Day

Because I am a human with my own intelligence and agency, I asked a simple follow up question (1st image).

The response completely undermines the whole premise of the question. Some research suggests you might need hundreds of plants per square meter to make any difference in air quality!

Let's try Google's new "AI Mode"... Right on the splash page, they suggest an interesting query: "houseplants that improve air quality and don't need much light". (1st image)

I went ahead and clicked it and was given a long reply with extensive citations (2nd image).

Very cool! .... right?

Yeah… Union politics make this kind of inevitable too.

This disclaimer, or a similar one, is present on all major LLM chat interfaces.

+1 for Kenji’s spatchcock turkey with lemon herb mayo. Easily the best and highest return to effort ratio on a turkey I’ve ever had

Altered States (1980): A professor obsessed with his research into the meaning of life pushes the envelope and nearly becomes permanently transfigured by his auto-experimentation. He realizes there is no answer he will find satisfying, finding meaning in his relationships with his wife and family.

In addition to the existing literature that discusses the prevalence of p-hacking in this context, we provide at least three behavioral reasons why it might be expected.

That’s consistent with what we are considering as a main factor in understanding how companies avoid p-hacking. Sounds like you have developed your own rule of thumb that is much less arbitrary than “p<0.05”, as continues to be the case in much of science.

Though this project is mainly concerned with p-hacking in e-commerce A/B testing, a close reading raises questions about academia as well.

Our results suggest p-hacking is not some inevitable characteristic of frequentist statistics, placing academic practices in stark contrast.

We analyze the distribution of p-values from over 2,200 A/B tests run by hundreds of e-commerce companies and find *little (null) evidence* of p-hacking.

We further use various counterfactual arguments to demonstrate why this should be considered a relatively "tight" null.

"p-hacking" is a widespread topic of intrigue and notoriety in science. We study the prevalence of p-hacking on a large online A/B testing platform.

Complementing related work, this study helps us understand how statistical criteria are used across different contexts.

💡I am happy to announce new research with Kartik Hosanagar. Available below and forthcoming in 𝘐𝘯𝘧𝘰𝘳𝘮𝘢𝘵𝘪𝘰𝘯 𝘚𝘺𝘴𝘵𝘦𝘮𝘴 𝘙𝘦𝘴𝘦𝘢𝘳𝘤𝘩.💡

Link (PDF): alexmiller.phd/research/p-h...

We study p-hacking, A/B testing, data-driven decision making, and more.

Short thread below👇

Putting a train down the middle of a freeway is one of those ideas that makes total sense at first but is actually really terrible and pernicious.

Its flaws are that it:
- prioritizes auto traffic
- is fundamentally hostile to human-sized transit modes
- discourages complementary land use patterns

This is terrifying.