For new PhD students in social science (and some old professors - I sure needed this a few years ago!): prepared some notes on what the "2024 PhD Tech Stack" should look like. Goal: your research is replicable, accessible, and efficient. 1/9 Replicable: whether or not the journal requires a full replication file, there is *no reason* every table and result in your paper can't be created with one line of code, and *no reason* you can't find every change you or a coauthor made. It's called version control! 2/9

Job market candidates: a quick thread on how the econ market works, incl. at business schools. 1) It is not unusual to get many 100s of applications. Nobody reads them all first-pass. Most places take a quick glance at letter/cv/intro, and keep the top x%. 2) You may think there are well-thought-out ex-ante fields schools will hire in. Sometimes this is true - but sometimes "general" ads are not actually general due to internal dept bargaining. You can't know - and sometimes the dept doesn't know yet - so don't worry about it.

It's Nobel Day, and the prize goes to Acemoglu, Johnson and Robinson! I believe AJR 2001 is the most cited economics paper of the 21st century (~20k!). A brief thread on why this literature was so influential, and slightly controversial, and my read on how AJR got here. 🧵1/x Briefly: "AJR polecon" posits 1) institutions, laws & culture persist 2) even if their origins can be for somewhat random reasons 3) leading to differential growth 4) with no 'end of history' liberal democracy trend 5) but societies where broad talent is used productively prosper

It's Friday and I'm feeling a bit feisty, so...I think a lot of innovation economists don't realize just how much influence UCL IIPP is having on policy, esp. in Europe. They are super connected politically and I don't think politicians realize how nonstandard their work is. 1/x IIPP is basically Mazzucato and friends. Almost every economist there is super heterodox. Ex: a housing policy paper commissioned by UK govt by them popped up on my feed. It was a long "it's not supply" argument. Here is the final post; look at who it is "of interest" to! 2/x

The hardest thing to explain to non-academics is how different technical rigor is across fields. Here's (yet another) article in PNAS, of course handled by Fiske, of course in psych, that we would have desk rejected. Paper is "Can Names Shape Facial Appearance?" 1/n The question is (and I have no idea why we care, to be honest) whether names match facial appearance, even if we go beyond gender/race/obv religion. Their headline result? Yes, but only for adults, so people "grow into" their faces. 2/n

A lot of bad AI policy is being driven by the fact that one side of the debate is sometimes really odd. "I don't believe exercise, Ozempic or calorie counting work, so I do a crazy fast every 3rd month" is literally the science of the most famous AI alignment guy. 1/3

https://twitter.com/ESYudkowsky/status/1816925777377788295

Many EA and alignment folks are really thoughtful and reasonable. But there is a blind spot in their separation of better knowledge (they are often smart!) vs. different preferences. Two ppl may disagree on policy b/c one has better info or b/c they have different goals. 2/3

Massive OpenResearch basic income papers are out (@smilleralert @dbroockman @evavivalt @AlexBartik @elizabethrds). Very much worth reading - my view is that it is an incredible RCT and an incredible disappointment. RCT was USD11400/yr for 3 years, 1k treatment, 2k control. 1/x The study was crazy, by the way. Very low attrition, time diaries, *blood draws to measure health*(?!) My favorite: they got Ilinois to pass a law that this RCT income wasn't taxable & didn't change other-benefit eligibility. That is, it was a net post-tax increase in income! 2/x

I came to U Toronto from Chicago today 10 yrs ago on Canada Day. So a short thread on Canada from an US perspective. And yes, living in Toronto is like living in NYC and talking about the US overall, but I've also been from Tuk to Tofino to rural Newfoundland & Quebec. 1/23 Best thing about Canada is the cultural emphasis on "be nice". Politeness is really the country's core value, very good for quality of life. I had a car in downtown Saskatoon stop for me to cross the road *when they had a green light*. 2/x

Many folks might be surprised at the low technical level of a lot of published academic research. A good example: snowball sampling. Get a small initial sample, ask for their friends/contacts, and build the sample from there. 178000 papers on Google Scholar using this. 1/4 In top 5 econ journals, the term has been used 7 times: either as a warning not to do it, or in the uncontroversial Trow/Coleman case where you are either measuring a dyadic variable (hence need both parts of the dyad) or are tracing out the size of a population. 2/4

Back from a great Paris conference last week after 4 visits to France in the last year - it's a very economically interesting country at the moment, with the caveat that it's not clear how much of this is Macron vs. long-run change. A short thread. First, the good: obvious cultural supremacy, the clear #2 after the US. Wemby, the Olympics, LVMH and the other fashion conglomerates, Lupin, Mbappe, Cannes, 2 straight World Cup finals, Djadja, Ernaux, Michelin cuisine, still huge and young Francophonie + the enviable lifestyle.

Toronto's President is at parliament today + injunction is coming to end encampment. I understand people have strong feelings on this issue & deep disagreements. I get it. War is awful. Intractable problems are intractable for a reason. I think we're close to resolution, though. That said, as our President mentioned, there have been 38 incidents reported to police. 6 of which he said would fit defn of hate crime. On day 1, both @mattgurney and I saw "Go back to Europe". A guy yelled about Hitler. The university has removed hateful posters almost daily.

A short note on the encampment that's now set up here in Toronto @UofT. I snuck into it today (I told them I was a fellow comrade who'd stepped out to get a cig - not exactly crack security). I think it's only fair to talk to the folks and see what's going on before judging. 1/x I did start recording video, just to ensure nothing was misinterpeted, when a fellow wanted to remove me from the site. I'm not trying to embarrass these kids, so won't post it, but suffice to say what I'm saying below is documented. 2/x

A few notes on Japan where I was last week. Quite fascinating economically. I have been many times over two decades rural and urban. 1) There are so so many foreign tourists. Regional Asian tourism + weak yen + collapse in tourism to China + was closed for multiple years. 1/x 2) Japan feels very cheap now, which it never did before. 1000 yen=$6.50. I would put costs overall at Southern Europe levels, aside from luxury hotels (see #1). Bottle of coke in corner store: $.80. Casual sit-down meal with tax & "tip": $8. Tokyo subway ride: $1.20. 2/x

In line with @tylercowen's interesting article on fertility, and the China news, I really think "people alive today will see peak worldwide population, and we have likely passed the peak absolute births per year ever in the 2010s" is a massively underrated economic phenomenon. Those of us in the US/Canada/Oz/NZ really misunderstand how much of the developed world has little, no, or negative population growth. *Every* developed country aside from Israel is below replacement fertility. *No* country has ever gone under then gone back up above.

NYT/OpenAI lawsuit completely misunderstands how LLMs work, and judges getting this wrong will do huge damage to AI. Basic point: LLMs DON'T "STORE" UNDERLYING TRAINING TEXT. It is impossible- the parameter size of GPT-3.5 or 4 is not enough to losslessly encode the training set. Briefly, how do LLMs work? They take huge training set of text (~the whole internet), then train attention model to predict next words following given user text. That is, if you say "The sun ", next words are likely "is", "rises", "emits". If "Hemingway's The Sun", likely "Also".

I'd go further than this. Reason economics is most successful social science is b/c of very high level of methodological rigor. Rigor, not hagiography/rhetoric, tells us what counts as logical argument or valid stats method. Other social sciences filled with "intuitive" nonsense.

https://twitter.com/ShengwuLi/status/1698737217990975890

When I talk to grad students about theory (& same is true of statistical inference), I point out that I've never written a theory paper where my initial conjecture was fully true. Good rhetorician can convince you of many things. A proper proof cuts through morass of imprecision.

One last word on "greedflation"-gate. My frustration with high-profile media coverage of out-there economics ideas comes down to fact that, unlike in science or math or many other fields, media coverage of econ has so little resemblance to the actual academic and policy debates. Certain ideas - MMT, some crypto, 'greedflation', goldbugs, Neobrandeis antitrust, some race/gender issues, some industrial policy - are 1) Very Online 2) have virulent defenders 3) usually (not always) associated w/ left politics 4) sold as "gatekeepers don't want you to know!"

Crazy stat: as far as I can see, Ontario (population 15m) has twice as many international students in two year colleges as the entire United States (pop. 333m). This is because "tuition" is a de facto low-wage work permit program in Canada (both pre- & post-grad). It's unethical

https://twitter.com/JohnPasalis/status/1666919274496270336

Even crazier: "public-private college" programs, where the 2 year degree is run by some fly-by-night storefront school but gets counted by CAN immigration as a "public degree" hence for work permit program (and then easy path to PR). If we want to auction off PR, just do that.

It's John Bates Clark Medal day - this year's winner is Gabriel Zucman from Berkeley/Paris School of Economics. I think you'll find this is somewhat of a controversial award, so let's start with the positives. Zucman's productivity is bonkers. Finished his PhD in Paris in 2013 (at the time *way* outside the "core" of the profession), then AP at LSE and Berkeley, multiple books, 18000 citations, 8 top 5s, running multiple institutes, tons of public speaking/consulting

A lot of misunderstanding about new papers on LLMs/generative AI and labor demand/wage (e.g., @danielrock and friends here arxiv.org/pdf/2303.10130…). Question: if tech is substitute for some tasks by labor type A, do total jobs doing A go up or down, and does wage go up or down? In the "farmer"/"steelworker" case: large productivity improvements lead to huge decline in labor demand and no change in relative wage. In the "calculations" case: large prod. improvements (from computers) lead to huge increase in labor demand and rise in relative wage.

Ok, let's go with one more insane AI application. As in the French app I showed the other day, I coded this in one evening, knowing basically nothing ex-ante about Flask or AJAX. This time, let's build your personal research index! Here's the idea - I want an LLM to be able any question I ask, in any form, about any of my own research papers, including those that aren't even online. I want to do it with <100 lines of code, for a couple cents per session, with no costly fine-tuning, and with high accuracy.