It’s that day of the week once more — when the birds are chirping, the suns out and spring is right here. Right here’s a ton of articles I’m dumping on you.
Anthropologists hint the break up between Homo sapiens and Neathenderals at Harvard.
Pedro Sant’Anna and Qi Xu have a forthcoming Journal of Econometrics on estimating difference-in-differences fashions when you may have compositional modifications. See Hong (2013) earlier for his personal work on that query too however utilized apparently to Napster.
Was simply testing the new Apple Imaginative and prescient Execs this week curious..
Tyler Random on easy methods to repair persistent peer overview issues.
Has a scientific tradition round citations had a optimistic or damaging affect on scientific progress? An older 2020 NBER asks one thing related.
The six proposed steps to assist AI assist science progress.
Peter Dolton and Richard Tol on the method of constructing a Nobel Prize in Economics. A bit style from the summary:
“We discover sturdy proof that the Nobel Prize rotates in a semi-regular means between the fields of economics. Earlier awards had been for a single paper, later ones for a physique of labor. Networks don’t matter, however having a Nobel pupil or co-author does. There may be some proof that the non-public preferences of Committee members had an impact on both area or particular person winner. The Committee’s selections modified after Lindbeck retired.”
This previous interview with Francine Blau stays one in all my favorites.
Stanford research on the risks of asking chatbots for recommendation. (Wish to see the research about asking people for recommendation as a management).
One other story of a person who fell right into a gap, and rapidly, after speaking extensively with a chatbot. Ended up investing 100k into some questionable enterprise concepts.
Google says we are able to change our gmail addresses.
The story of how Anthropic unintentionally posted the supply code for Claude Code on-line will probably be a narrative advised for a really very long time. It isn’t catastrophic for Anthopic, however relatively is one thing extra like an “operational blueprint of how the present model of Claude Code is designed to work.”
Due to the Claude Code supply code gettin posted on-line, we now know there are 187 spinner verbs. See what number of you acknowledge.
Wes Bos@wesbos
Claude Code leaked their supply map, successfully supplying you with a glance into the codebase.
I instantly went for the one factor that mattered: spinner verbs
There are 187
12:37 PM · Mar 31, 2026 · 2.17M Views
707 Replies · 1.89K Reposts · 26.7K Likes
However right here’s one other breakdown of what was discovered within the leak. A bunch of attention-grabbing options had been flagged. Right here they’re. The largest reveal was an unreleased autonomous agent mode referred to as KAIROS. It has scaffolding for a talent referred to as /dream that does “nightly reminiscence distillation” with each day append-only logs, Github webhook subscriptions, background staff, and cron-scheduled refreshes eery 5 minutes. It’s an always-on, background operating agent, and in keeping with the writer, it’s actually important produce leak. One other one which me was a regex sample that detected whenever you had been annoyed by what number of f-bombs you dropped.
Peter McCrory, head of economics at Anthropic, has issued a new report on using Claude in Australia as a part of his bigger work creating an financial index for AI.
Can’t wait to see all the brand new films and footage of the moon. A few of them are filmed within the sort of tools used for Apple Imaginative and prescient Professional immersive films.
On the Artemis II clothes patch.
New Anton Korinek and Joseph Stiglitz NBER working paper on “steering technological progress”. From the summary, “This paper asks easy methods to information revolutionary efforts in order to extend labor demand and create better-paying jobs whereas additionally evaluating the restrictions of such an method.” Wish to guess me that none of that occurs?
A brand new Apple macOS patch appears designed for Claude code safety issues as a result of it stops you from pasting one thing into terminal you bought from the net. Who was doing issues like that earlier than CLI AI Brokers who didn’t already know what they had been doing?
A brand new “many analyst design” paper utilizing AI brokers.
The sudden surge in Claude customers is inserting congestion on the platform and slowing use for everybody else. Surge pricing coming?
Tips about organising “Tasks” with Claude Cowork. However in all seriousness, I can’t for the lifetime of me work out how that is any completely different from how I take advantage of Claude Code. Each time I learn one thing by somebody about Cowork, I simply suppose this feels like what I already do with Code.
How Marcus Sensible helps the Lakers win.
The largest live shows coming to NYC subsequent week. I’ll be there talking on the NABE convention, so I used to be trying round for stuff to do.
I noticed Carlos Cinelli current this final semester right here at Harvard; now it’s out in Evaluate of Economics and Statistics together with an influence home group of coauthors. This continues Carlos’s work on sensitivity evaluation and omitted variable bias. I believe this paper may can help you lengthen the unique Cinelli and Hazlett work to extra normal remedy results parameter just like the ATT. I believe, however could also be mistaken, that most of the earlier work on this like Emily Oster’s methodology (but additionally Cinelli and Hazlett) is designed for regressions with a single remedy indicator. However beneath remedy impact heterogeneity, the ATT have to be estimated utilizing one thing just like the Oaxaca-Blinder methodology. See right here Tymon’s Restat. The brand new Restat by Cinelli, et al., although, I believe let’s you utilize causal ML strategies to get quite a lot of parameters, together with the ATT, however they might nonetheless require frequent assist as there seems to be a propensity rating beneath the hood. The issue is that the saturated regression wanted for the ATT is a single remedy, and it’s a collection of remedy interactions.
Listed here are two ancestors to their paper although. The Oster bounds paper. And the Cinelli and Hazlett paper. After which one of many ones that began all of it, the Altonji, Elder and Taber JPE. I’m most likely imagined to now hint it again to Leamer, however I’m going to cease there.
The APE Mission is sort of at their purpose of 1,000 AI generated papers. As of this writing, it’s 942.
Judea Pearl’s hierarchy and foundations of causal inference.
The private essay industrial complicated. Kind of suppose you need to pay a tax should you name one thing the “[thing you’re criticizing] industrial complicated” though it’s form of enjoyable to provide you with a brand new model of 1 if I’m being sincere.
Rambachan and Roth had a JASA ;final yr on design primarily based uncertainty with quasi experimental work.
What’s the impact of immigration on older grownup mortality? NBER working paper. However right here’s a writeup by it within the Harvard Gazette.
Has self protection grow to be captured?
A new article forthcoming within the Journal of Marriage and Household from Harvard Enterprise College researchers seems at division of labor within the household and corresponding resentment and gratitude. The core discovering is in regards to the hole between companions’ expectations (not simply their conduct) and the way that hole over time accumulates into resentment or gratitude. The discovering has to do with asymmetries between women and men. Some ladies had been grateful for males’s participation in bodily labor, whereas others anticipated important bodily and cognitive contributions and resented their absence. Most males anticipated to contribute bodily family labor and felt grateful when ladies’s unpaid work freed up time for his or her paid work. Males’s resentment emerged after they felt their bodily contributions had been underappreciated, or after they felt their companions had unreasonable expectations for his or her cognitive labor participation. The cognitive labor stuff was attention-grabbing as a result of it included issues like remembering appointments, planning and anticipating wants. The article means that the fairness issues are literally there — within the cognitive labor. That’s the place resentment builds and could be very onerous to do away with when it exists and has accrued. They discovered that it tended to say no, although, when males elevated their attunement and ladies their use of delegation.
An interview with Shane Greenstein at Harvard Enterprise College. Location nonetheless issues.
Right here’s a wellness piece summarizing the idea of the nervous system for a normal viewers. The core concept is that we’re disregulated when our menace/combat/flight mode (“sympathetic”) and our relaxation/security (“parasympathetic”) mode are out of stability, usually as a consequence of persistent stress or trauma. Our physique will then fireplace menace responses even when there are none. Dysregulation, apparently, isn’t normally from a single traumatic occasion, however relatively, a bunch of small “t” traumatic occasions that principally add up. Additive remedy results, in different phrases, form of just like the resentment and gratitude piece the place issues accumulate. Issues like studying as a child to suppress your feelings to maintain the peace, or taking over different folks’s issues as a result of the adults round you wanted you to. These coping patterns get wired in, after which many years later your nervous system remains to be operating them as defaults even when you do not want them anymore. However the purpose isn’t actually calm, apparently. The purpose is self acceptance, self respect, accepting all elements of your self, loving all elements of your self, even the sloppy and soiled elements.
An extended profile on Harvard’s President, Alan Garber. Garver did his PhD in economics at Harvard and an MD collectively at Stanford, touring coast to coast. I don’t even see how that’s humanly doable. The mental biography is attention-grabbing. He began in biochemistry, obtained pulled into economics by a Dunster Home tutor and, like so many, by taking Ec 10. He grew to become a well being economist who utilized causal instruments to issues like antibiotic resistance and heart problems. Spent 25 years at Stanford earlier than coming again to Harvard as provost in 2011, then grew to become president.
We hear about how AI goes to remodel work, possibly even eliminating it, but when that occurs, then what occurs to public finance? An NBER working paper.
Worker #8 nonetheless works at Apple — the place he’s labored ever since he was 14. As soon as layoffs had been taking place and he was spared as a result of they advised him his severance package deal would’ve been too giant. Woz gave him 5700 of his shares on the firm’s founding, which might now be value one thing like $140 million. New York Instances has a chunk about him.
A new paper by Brodeur and 350+ coauthors in Nature presents one of many largest reproducibility initiatives in economics and political science up to now.
I4R@I4Replication
🧵1/ Our first meta-science paper (with 350+ coauthors) is printed right this moment in Nature. It presents one of many largest-ever reproducibility initiatives in economics & political science.
Right here’s what we discovered 👇
3:12 PM · Apr 1, 2026 · 96.5K Views
10 Replies · 222 Reposts · 567 Likes
Apple hasn’t stopped investing assets in Apple Imaginative and prescient Professional. Some new applied sciences could assist with rendering much more complicated graphics.
Resort Villa Favorita in San Sebastián.
One of many issues I’ve been doing is making an attempt to create an government assistant, like Chris Blattman, however doing it myself. I’ve been doing it with Obsidian, and to this point it’s going okay. Although I put it down. Listed here are some issues about Obsidian dashboards. Right here’s extra about making stay dashboards on it. Claude had instructed this was the very best for what I wanted.
Did I ever publish this beautiful pessimistic idea paper by Acemoglu, et al. about AI inflicting a collapse in collective data? I’ll be sincere — that’s sort of the way in which I’ve been excited about it too. However I are likely to have fashions which are very partial equilibrium in my thoughts that function warnings, not a lot as descriptions of a actuality coming. However in a means so are his. He pushes for extra regulation of technological innovation in order that progress can proceed and wealth be extra equitably distributed.
However Andrew Peterson has additionally written on data collapse and AI.
A number of traumatic occasions from childhood discovered to be linked to addictive behaviors. I believe we knew this, and possibly this text doesn’t per se budge lots of people’s priors, however I assume I noticed this text and was .
Gen AI boosts productiveness, however it can’t flip novices into consultants. That’s an attention-grabbing means of placing it. And I believe since all of us are feeling like Claude Code is pushing us past our personal manufacturing risk frontier, not simply on time, however on duties we thought we might by no means be taught, it’s most likely value printing that out and pinning it to the wall. I nonetheless suppose the purpose must be our personal private development and accelerating to our furthest potential with AI (and occasional, and tv reveals, and lengthy walks, and watching Luka on YouTube, and kitties, and so forth).
Do you have to subscribe to your buddy’s substack? The Ethicist is right here to reply that query for you.
The perfect bidet in your bathroom in keeping with NYT. If I used to be wealthy, as an alternative of donating cash to charities, I’d ship them these Japanese bathrooms.
I’m going to speak about this subsequent week however do you know that there’s an AER remark by Kranz and Pütz that the Brodeur, et al. authentic research in AER had not adequately handled the rounding downside related to extracting coefficients and commonplace errors. Brodeur, et al. replied. So get this — the unique discovering that diff-in-diff papers confirmed indicators of p-hacking really wasn’t sturdy (the IV consequence was sturdy although). The discovering for diff-in-diff was mockingly the identical “rounding” downside that I discovered final week with my preliminary extraction at 651 papers. However I’m going to copy what they did subsequent week to the APE pull I did and see should you can right it.
Vibe coding has prompted Apple’s App Retailer to begin delaying approval of apps. I assumed that was attention-grabbing as a result of that’s most likely just like the refereeing / journaling situation I’ve written about earlier than with automated papers.
And with that, I’m calling it quits. Want me luck. I’m going to faux the climate isn’t beautiful right this moment right here in Boston and see if I can get numerous work accomplished.
