Regression is a tool for making comparisons If you don't know / can't easily explain what comparisons you're trying to make, then you don't understand the regression you're running

It is more rhetorically than epistemically useful, to “boil down to one argument” a view that one disagrees with.

These charts are helpful for communicating some basic facts, but they don’t (without a lot of extra work) tell us much about needed policy reforms. The basic shape of the chart will look the same even if we make dramatic changes in policy.

Management matters, even in the public sector when they cannot control payroll or even easily fire anyone. A manager one standard deviation better than average raises productivity by 10%. Yes, really, managerial pay is justified. 1/

It’s easy to underestimate the importance of extremely fast and convenient EV charging… EV batteries wouldn’t need the same total capacity so cars can become lighter and more affordable. 🚀

Perhaps the greatest failure of the economics profession has been our inability to sell the idea that we should tax "bads" like congestion and pollution. One reason building is so expensive is that we have so many regulations instead of just prices.

Lab leak in Spain… the world has done shockingly little to improve biolab safety.

Why North America has a housing crisis, in one tweet.

Policy wonks views about “epistemically rigorous evidence” may not be politically potent. But if the benefits are broad-based and large enough to be noticeable to the public, then good policy is likely to become more popular post-adoption.

Asking the right forecasting questions is neglected for lack of agreed-upon ways to credit a good question (Cf. Philip E. Tetlock) Does the rise of AI-forecasting-agents make question-posing even more important? Yes and no - the nature of the question-posing challenge will change.

Can we predict the past? 🔮 New open-access paper in American Historical Review proposes 'Retrodiction'—using gaps in the archival record to test historical theories. Interdisciplinary collab feat. David Gill, Marc Trachtenberg, Philip E. Tetlock Cendri Hutcherson, and more 🧵👇

Faster, better, and stronger than ever 🇨🇦

.Rodney Brooks has just published his annual predictions scorecard about AI, robotics and technology more broadly. The best part is that he's been doing this for 8 years and scrupulously checks on the accuracy of his past predictions. A must-read.

I think it’s really unfortunate that articles in the natural sciences are so short and provide relatively little detail about the “work”/how of research. Harms knowledge transfer and reproducibility. I hope AI raises expectations for all sciences in terms of how much

I like this, but I’m happy to say the part of EconTwitter I follow is mostly posting about how great AI is and how to get the most out of it!

Personal news: I'm leaving my current startup role, looking to figure out what's next. I'm interested in making AI go well, and open to a variety of options for doing so. I have 10+ years of quant research and technical management experience, based in London. DM if interested.

Bad: small samples of undergrads Also bad: large samples of people (or bots) who will just click the closest box.

This is great, but does anyone have empirical evidence of how much this actually helps LLMs?