Good morning Vietnam! That’s a Robin Williams joke. But seriously good morning. I’m super excited to somehow manage to pull a ton of links off of four different devices and share with you thinks I’ve been doing and reading this week, and also not doing and not reading.
I discovered Detroiters recently, starring Tim Robinson and Sam Richardson. Netflix started streaming a week or two ago after a while of only being available in random streaming services. Probably the funnest show I’ve seen in a million years. Well maybe they’ll get a third season. I need to find out how the people at Devereaux Wigs are doing so I hope so.
London School of Economics is hiring an history of thought and economic history. Listen to this description:
“We are particularly interested in candidates whose research connects the history of economic ideas to the economic history of the same time and place, including the impact of economic ideas on policies and practical changes/outcomes within economies, and/or the effect of historical economic change on economic thought.”
Swoon. In another life, I would be applying for that. Here is Mary Morgan’s wonderful History of Econometrics.
Sung Jae Jun and Sokbae Lee are estimating the effect of persuasion on the persuaded parameter using diff in diff. HT Pedro Sant’Anna.
This AI paper on intelligent systems updating their beliefs in response to external unexpected outcomes — reflection — looks interesting.
Kaiser and Shapiro on the demand for water quality and the Clean Air Act has an event study equation I have promised someone to figure out. Always takes me work to decipher event studies in a regression where all the coefficients are plotted (ie no clear dropping of a baseline indicator), but I’m confident I’ll figure it out.
Surgeon general announces we are in a global epidemic of loneliness. Here’s the Chinese version of Deaths of Despair by Anne academia and Angus Deaton.
NYT reported a young man’s family has sued characteristics.ai after his unexpected death by suicide. Records show he had developed a strong emotional and romantic attachment with his chatbot.
Another AI company with chatbots used for intimacy and fantasies had a data breach — sort of Ashley Madison type of thing as it appears to have been hacker related. I expect this LLM-human interaction to be something that becomes more and more prevalent.
Teenager entrepreneurs making five figures as quasi-influencer things. Don’t forget to study kids!
HBO Max’s Batman spinoff, the Penguin, continues to be streaming services major hit. Try as you can, you will not see Colin Farrell beneath those prosthetics. I cannot believe it’s him. Great show.
Jon Roth at Brown says you may be interpreting your event study coefficients incorrectly — and displaying the wrong ones (wrong ones you say? Yes he says). Friends don’t let friends use the short-gap calculation. Highly encourage you to read this short and pithy 9-pager.
Susan Athey, Keyon Vafa and David Blei write a follow up to their “foundation model” paper on CAREER that they trained with 23m online resumes to predict labor market outcomes, which bested the next best method by around a 17% improvement. This time CAREER is used for estimating causal effects with respect to wage disparities in a modification to the Oaxaca-Blinder-Kitagawa style decomposition.
I gave the MCU another chance with the new Disney+ show Agatha All Along. Not bad. But your mileage may vary. Never thought I’d see the day, though, where I was this aloof to superheroes. Young Scott would be so disappointed at how bad, he’d think, adulthood must be. Jokes on him. Just wait until he learns he can’t eat whatever he wants any more.
What books and texts has ChatGPT/GPT-4 trained on? A lot of stuff including copywritten material. The things it’s memorized is basically a function of the number of times those things are mentioned online, not surprisingly.
Whats the one personality trait, “according to psychologists”, that predicts divorce in couples? Neuroticism defined as high reactivity and emotional instability in response to every day stress.
“Neuroticism, characterized by emotional instability and high reactivity, is a key driver of divorce for many couples. While all relationships experience ups and downs, individuals with high levels of neuroticism are more prone to interpreting those ups and downs in negative and damaging ways.”
Psychologists probably say outlandishly outlier levels of stress are probably also high predictors too, but I didn’t find that one.
People are raving about Anna Kendrick’s directorial debut, Woman of the Hour, which has 92% on rotten tomatoes and is streaming on Netflix. I started and quit; it’s about a true story serial killer who also appeared on an old dating game style game show several decades ago. It was too much for me to watch but maybe you’ll want to finish it. I’m noticing these days I’m sensitive to this material.
Prompt engineering advice to personal success. On a different note, I hate that phrase “prompt engineering”. I just talk normally with and to ChatGPT-4o. Theres no need in my experience to any longer devote so much advice column stuff to how to communicate like a regular person to the LLM. Just be clear and ask it in different ways if you get stuck and always flip back and forth with your code between ChatGPT and Claude. It’s not rocket science.
Anton Korinek wrote a second article for JEL on AI and LLMs for economists when the first one has only been out a short while.
Our electionGPT simulation using future narratives and Monte Carlos continues to bound Harris’s number of electoral colleges votes at around 275 lower bound to 300 upper bound. Shiny app for those interested is here. Day to day variation after reading 100 newspaper article summaries and 100 repeated trials still shows the variability coming from the swing states interestingly. You can click down on the shiny under National to see the predictions without any embedding of newspaper articles and then with to see the updating that happens. We were increasing the sample size with both more articles read daily (around 175) and more of each article (the entire article not simple summaries), as well as about 10x more trials, but it’s been hard to get the LLMs to consistently finish. Hopefully will have something soon. But until then, here is the mean prediction for our different “journalist voices” since we started this in August that we did a couple weeks ago. It’s the distribution of all daily (average over all 100 trials) electoral colleges votes.
If you haven’t had a chance to listen to Miikka’s podcast interview that I posted this week, check it out. Miikka was an Angrist student who started out at Columbia after graduation then went to Amazon. Good story for learning more about how both halves live.
Sam the 12yo Bassett hound in north dallas that I had my eye on got snatched up before I could go up there and convince him to give me a chance. But there’s some new faces there now so all is not lost. Bruno looks like a good boy, and he’s potty trained. No mention of what he thinks about neurotic cats with emotional instability and high negative reactivity so I may have to just email about that.
Top economics departments by Repec ranking for those who want to know. No real surprises there.
Speaking of top economics departments, I managed to get money from our graduate school to start recruiting for our masters program in economics recently and helped organize a recruiting dinner for Baylor undergrads. We invited three alums — a 2014 grad who’s since gone on to a productive career in data science, a 2017 grad who has since gone on to a productive career in AI and consulting, and a 2023 grad who started out at Deloitte and has come back to Baylor to work in finance administration. I convinced my colleagues to have Torchy’s Tacos cater the event, took home a ton of tacos, and then immediately asked my ex-wife if she’s please take them so I wouldn’t eat them. We had a presentation about the program by my colleague, Finley Edwards, who runs the masters program and then I interviewed our three alums on a panel. It went great. I came away convinced that the masters program has value added for the right kind of student. If you’re wanting a rigorous graduate degree where you can learn from a great, research active faculty, and want to gain considerably more quantitative methodological training, I think ours is the masters to consider. It just became very clear after interviewing our alums that the careers they are in would not have been possible with only a bachelors. And yes I know that I don’t know the counterfactual. One of the things I was not expecting was when I asked them about the course work, and they all said how incredibly hard all the classes were, but that graduating and surviving had taught them a lot about themselves and their own potential — I just wasn’t expecting to hear that the coursework had brought upon personal transformation. Beautiful and good things often are the most difficult things too. It reminded me that holding our students to high standards and high expectations but believing in them to be successful and working with them to be successful is an important goal for us as educators. The masters in economics at Baylor can be completed in 30 hours and I also learned that students have found the ability to have smaller classes and work directly with the faculty was a very important part of the program’s value for them. I encourage you to reach out if you’re curious about the program, and apply here.
Pivoting back to Detroiters, good news — Tim and Sam are best friends in real life too. But what I really want to know is if one of them were to die, would the other one die no matter where in town the other one was because their hearts would just know it?
Susan Athey, Stefan Wager and Jann Spiess videos on applied machine learning and causal inference from Stanford.
Robert Browning’s famous first letter to Elizabeth Barrett started with “I love your verses with all my heart.” I quoted it in my revision to the mixtape yesterday in saying how much I enjoy reading Guido Imbens’ work but may take it out. Gonna sleep on it. I’m getting there on the revision. Wish me luck. Five more weeks to deadline!
What happens if you try to make your LLMs anxious? They become more biased.
“Anxiety-induction not only influences LLMs’ scores on an anxiety questionnaire but also influences their behavior in a previously-established benchmark measuring biases1 such as racism and ageism. Importantly, greater anxiety-inducing text leads to stronger increases in biases, suggesting that how anxiously a prompt is communicated to large language models has a strong influence on their behavior in applied settings.”
I say please and thank you when I talk to ChatGPT.
Manicode lets you interact with your codebase and terminal using NLP.
A free book on AI safety and ethics.
The current bid on this 1987 240 Volvo DL wagon is $8000 with 2 days left. Highly encourage you to consider getting one. I own a 1991 240 Volvo wagon and the other day I was driving it and passed a guy who leaned out of his driver side window and screamed at me as loud as he could “Turbo brick!” as loud as he could. So think about it.
Longrun effects of fluoride on health and economic outcomes in the Journal of Health Economics.
The late, great Rodney Andrews gets a park named after him in his hometown. A truly beautiful man. He passed away around a year and a half ago.
I may have already shared this, but here’s a new JPE on railroads and economic growth.
And that’s a wrap! I accidentally fell asleep on the couch last night and woke up early — too early. So I may try to go catch some zzzz’s. Have a great Saturday!