Best of #econtwitter - Week of December 4, 2022 [1/3]
Welcome readers old and new to this week’s edition of Best of Econtwitter. Please submit suggestions — very much including your own work! — over email or on Twitter @just_economics.
This is part one of three.
Paper summaries
Park, Leahey, and Funk (2022) pull together a bunch of interesting data to document a decline in how disruptive science is. E.g., here is the share of new pairs of words in academic titles, relative to the total number of pairs. Titles increasingly recycle old pairs of words.
And here is the average score of a kind of disruption index; it's based on whether papers that cite you continue to cite the papers you reference, or if they stop citing your forebears when they cite you. The latter case is judged more disruptive. Average disruption is falling!
🚨Wake up babe, new study of market-rate housing just published🚨 (Spoiler: it's a banger) 🧵
Xiaodi Li found that when a new luxury apartment is built in NYC:
📉 nearby house prices fall
📉 nearby rents fall
🧑🍳more restaurants open up &
⚖️low-rent units do not see rent hikes
1/n Sharp Econometrics: A 🧵
Chistopher Sims gave a talk back in August (mediatheque.lindau-nobel.org/recordings/402…) in which he provided a thoughtful rejoinder to the "Mostly Harmless Econometrics."
AP's textbook captures some of the lessons of the 'credibility revolution.'
Worked through a lovely paper that gave me a mild crisis my understanding of OLS, so I naturally must inflict it on the rest of you [and ask a question].
The paper is "Oaxaca-Blinder as a Reweighting Estimator"
eml.berkeley.edu/~pkline/papers…
Goal: estimate the ATT using OLS.
Have you ever wondered how proximity affects knowledge sharing in a co-working space? A recent study by Roche, Oettl, and Catalini looked into this exact question.
The above thread was entirely written by ChatGPT. I fed it, as a prompt, the following excerpt from my own website. (Links to everything below)
^ChatGPT is the new OpenAI chatbot that blew up Twitter the past few days. The UI is very simple and the results speak for themselves
ChatGPT pop-up section
This section is going to break the gmail web length limit and you’ll have to open this page in Substack to see all the tweets!
I have helped run an AI-based entrepreneurship program for years, written papers on the econ of AI, and follow the field quite closely. Nonetheless, I am *shocked* by how good OpenAI's new chat (chat.openai.com/chat) is. E.g., you can no longer give take-home exams/homework.
Let's go even crazier. Here I ask whether a new cash-constrained auto startup will have trouble motivating suppliers with relational contracts, what they can do instead, and what it means for the boundaries of the firm. Heck, this might be an A.
2 implications: 1) you can't give take-home essays/ assignments anymore. 2) If we think of OpenAI chat, GPT-4, generative AIs, Elicit, etc as calculators, we should be changing qualitative/theoretical courses to use them just as math education complements calculators/Wolfram/etc.
The latest GPT-3 model is very impressive. Have fun writing take home exam questions for your undergrads.
To follow @orgRem @pierre_azoulay and @Afinetheorem it seems like #ChatGPT can be great for talking points at your academic conference.
This is #ChatGPT responding to "What are some open questions in the field of innovation research."
@JonSteinsson While I respect your work @JonSteinsson I'm not sure you've ever written anything that compares to this.
I guess GPT-3 is old news, but playing with OpenAI’s new chatbot is mindblowing. chat.openai.com/chat
We’re witnessing the death of the college essay in realtime. Here’s the response to a prompt from one of my 200-level history classes at Amherst
Solid A- work in 10 seconds
Hey grad students in economic theory, stop hiding behind the math and actually apply game theory to real world problems.
1/
Compare: (a) original title & abstract to (b) ChatGPT shortened abstract & then ChatGPT generated title based on abstract it shorted. Poll below
My most cancelable take is that GPT-3 will usher in a new golden age of proctored timed exams for big undergrad intro courses (esp STEM), and despite their initial apprehension, the median student and median instructor will find themselves much better off under this new eqm.
Meta takes on ChatGPT
I feel like every month these days there’s some fairly startling advance in AI and there’s zero impact on politics or policy.
The NYT still has 0 mentions of ChatGPT, which is basically all tech has been talking about for the last few days.
Instead, 3 out of the top 4 articles in its tech section are about Elon and Twitter - mostly negative, of course.
^Tyler Cowen: “More than any other time, if you are not on Twitter, you just don’t know what is going on.” If you are not on twitter, or subscribed to this newsletter :)
My general critique of AI language models is the same critique one should have of models in general, which is that giving an approximately correct solution in 98% of cases isn't particularly valuable if the 98% are trivial and the 2% are what you really care about.
^related. But only need to be 100% as right as the xth-percentile human — not 100% right overall
