Ingar Haaland

Congratulations to Stefanie Stantcheva (@s-stantcheva.bsky.social) of @harvard.edu, winner of the 2025 John Bates Clark Medal! #econsky www.aeaweb.org/about-aea/ho...

It managed to get the details better when asked to fix the awkward elements, though it looks much less like me

It's not perfect, but I'm pretty impressed at 4o's first take of me posing at a beach with a beach ball (same shirt as in the reference photo).

Democracy dies in deference

I'm surprised you're not a fan. Overleaf allows for a *so much more* dynamic workflow than Word. Manually typing point values and p-values in a draft instead of using macros is almost guaranteed to produce errors (and lots of frustration).

Agree!

I don't think one is "better" than the other, traditional labs have many problems too. Plain vanilla experiments using convenience samples from "online labor markets" might become less relevant, but those are only a small subset of the broader category of online experiments

Yes, in general, if you want to engineer a null, just make the experiment complicated

Hello, Bluesky! ☀️

We fund frontier research in Europe—bold ideas, unexpected discoveries and science that shapes the future. So it’s only fitting we’ve landed here. Sorry for being late.

Follow us for updates on ERC funding, research policy, and our grantees' discoveries.

Screening performance and characteristics of breast cancer detected in the Mammography Screening with Artificial Intelligence trial (MASAI): a randomised, controlled, parallel-group, non-inferiority, ... The findings suggest that AI contributes to the early detection of clinically relevant breast cancer and reduces screen-reading workload without increasing false positives.

New: The largest medical A.I. randomized controlled trial yet performed, enrolling >100,000 women undergoing mammography screening
The use of AI led to 29% higher detection of cancer, no increase of false positives, and reduced workload compared with radiologists w/o AI thelancet.com/journals/lan...

The new OpenAI model announced today is quite wild. It is essentially Google's Deep Research idea with multistep reasoning, web search, *and* the o3 model underneath (as far as I know). It sometimes takes a half hour to answer. Let me show you an example. 1/x

I find it funny how these amateurs at OpenAI are basically begging us, their paying subscribers, to test out the competition, especially when the competition is *free* with *unlimited requests*

In my BSc course, "Incentives, Politics, and Behavior," I am introducing students to interesting topics at the research frontier and making them think critically about causal identification. Today, we're doing "From Extreme to Mainstream: The Erosion of Social Norms," which feels almost too relevant

After you've been "used" to o1, 4o feels like a degraded experience. It's much more *intelligent*: better answers, asks for clarifications when needed, and follows instructions. It might be overkill for some tasks, but I've started getting annoyed when I'm out of o1 requests

An interesting aspect of o1 vs. 4o is that the former *really* pays attention to your custom instructions, whereas 4o mostly doesn't care. Ask for short, pointed responses and o1 will deliver; 4o mostly doesn't notice. Important to change "legacy" instructions created for 4o

Priceless take

**New working paper**

How does the under-representation of females in Economics affect the career trajectory of female Ph.D. students?

Sahar Parsa and I look at this in a new working paper by exploring sabbatical leaves taken by female professors at top-50 US Econ departments.

Understanding Economic Behavior Using Open-ended Survey Data - Social Economics Lab We survey the recent literature in economics measuring what is on top of people’s minds using open-ended questions. We first provide an overview of studies in political economy, macroeconomics, financ...

How can open-ended survey questions reveal deeper insights into economic behavior? Our completely revised paper (with new title!) explores recent developments, practical applications & best practices for designing & analyzing these powerful questions. Read more socialeconomicslab.org/research/wor...

Jeff Bezos never wanted this cartoon to become public.

He killed it, and as a result, pulitzer prize editorial cartoonist Ann Telnaes quit.

Make sure everyone sees this cartoon.

Congrats 🎉

You can just make a public version of your Overleaf document instead

Want to add a few questions to the General Social Survey? The 2026 module competition is open now! gss.norc.org/content/dam/...

New paper: We combine evidence from a survey (correlational), bank spending data, and online experiments to show that job loss increases risk taking.

Joint work with @abbysussman.bsky.social, Carlos Vazquez-Hernandez, Daniel O'Leary, and Jennifer Trueblood

What just happened A transformative month rewrites the capabilities of AI

Bluesky can be a fraught place to post about AI but it is worth noting that the buzz over o1 (& now o3) is not “hype.” We know o1 can actually do some very hard tasks (see my post) & o3 appears to represent a big further leap.

They aren’t AGI, but will matter. www.oneusefulthing.org/p/what-just-...

Alex, you aren't on the free tier?

Claude Fights Back ...

Claude Fights Back, crazy experiment www.astralcodexten.com/p/claude-fig...

I find it pretty impressive that is easily handles requests like this (just tried copying in the literal example of a flowchart from wikpedia)

Perhaps I'm a simple person but all these use cases as well as a ton of others work excellent for me

Don't forget TikZ figures! Or ask it to clean up your LaTeX preamble! Or how you can easily include multiple screenshots using a loop! If this isn't black magic what is

🤔