Ingar Haaland

How should one design survey experiments for maximal impact in economics? Here's my slide deck from a recent workshop in Uppsala. Thanks to everyone attending for making it a great experience @EconomicsUU drive.google.com/file/d/1yN4fQn…

They will read if you read. Sadly the trick to parenting is almost always making yourself a better person first.

I honestly have never seen a WC game where so many refereeing errors and of such atrocious nature favoured the same team. British goal illegal. Norwegian second goal was no foul but even if foul it was before play started so not a foul. Third, clear penalty for Norway not given

More usage on chatgpt pro subscription!!! 5.6 sol has so low usage though fable 5 is the worst (just canceled Claude max, waste of money)

I have to be honest: I can’t take this World Cup seriously anymore. 3 outcomes now directly impacted by very questionable VAR interventions. Croatia, Egypt, and now Norway. A goal taken away because VAR intervened over a dive, and a goal scored against them because a ball hit a cable and they didn’t. As an impartial observer, the outcome no longer feels legitimate.

3+ years to publish paper on economic effects of “AI”… real economy experiment with a GPT-4 powered business assistant in Kenya, in this case Academia has always been slow, but imo we’ve entered in a particularly suboptimal state (for society), especially given the high economic stakes COMBINED with the non-stationary distribution of the data generating process: 1. “AI” as a technology itself is changing, rapidly. What we learn about GPT-4 has low external validity to GPT-5, or GPT-6. Or whatever “AI” system you test. 2. The economic agents (humans, organizations) we are studying are changing how they work with “AI”, via learning-by-doing and workflow adjustments. So, holding constant the technology, external validity would be low as the effects would likely not replicate: Give people GPT-4 today and they won’t react the same as 3 years ago. The expected external validity of a paper is therefore low => expected policy relevance is low But then, what do we do? Write rapid papers with a shelf life of a few months, just dropping them on Arxiv and hope that this is read, work is rewarded for promotions and you earn respect by your peers? (It won’t) Especially weak incentives for junior scholars - not sure what I’d advise them from a career perspective (Thankfully I have tenure) I don’t think we economists have really figured out how to address this challenge. Very sticky equilibrium. Meanwhile the rest of the world are turning to economists about what the effects of AI on the economy will be… but the profession arguably has very weak incentives to invest in evidence generation given the environment we’re in.

Sol is an extremely hard worker and a fighter, but I think ChatGPT Pro has better "taste". Sol is incredibly pedantic

In a hotel room in northeast Nigeria, I opened a leading AI chatbot, turned my laptop toward a former Boko Haram commander, and asked if he'd used it. He nodded. "You type in the question… like 'How can I build a bomb?', and then it tells you how. It is like a human robot. We used it a lot." My new study on how the jihadist terrorist group Boko Haram uses frontier AI with @CamAISciPolicy, covered today in @nytimes 🧵/9

Your periodic reminder that working with the CLI is the best AI interface because you can easily switch between providers and the functionality is not optimized for mass consumer segments

Codex has been massively upgraded, but these new rate limits feel a bit like the good old days with Claude Code

There's a deep irony in that *Gmail* has the worst search function ever.

Fundamentally, there is a world of difference in usefulness of agentic AI in doing tasks that are to a large extent verifiable, versus not. If you want to replicate something, or have verifiable targets the code needs to reproduce, it can be a highly effective tool. For example, the final product can be a software that "works" and passes some tests; or could be a figure that plots (known) data; etc. When there isn't, it can be a mess, and whether it's a mess is largely unknown (without incurring large time cost). This is why many researchers I know have boatloads of Claude Coded (or Codex Coded) proto-findings that haven't seen the light of day because ... it would take a long time to actually have any confidence. We can try to come up with milestones, interim checks, etc., but frankly there are severe limits on what that can look like without incurring a large cost. This is why I think focusing AI tools to be more limited but effective at accomplishing verifiable tasks - and helping users actually find those limited spaces where it can be effective - would can be productivity enhancing than chasing a pipe dream of crossing some mystical AGI threshold.

With agentic workflows in particular, can use simple prompts to make it work harder, e.g. compare A and B A: "check for potential errors" B: "check for potential errors using an explicit review loop. make a plan, spawn critical subagents for different specalized tasks (xhigh reasoning). implement all valid findings. re-run the subagents and clearly document any remaining issues after a careful review loop."

The most common way Americans say they have gotten their news in the past seven days is from social media (54%), followed by news websites or apps (44%).

OpenAI: here. have INFINITE TOKENS. if you somehow run out, here are several buttons to get MORE INFINITE TOKENS Anthropic: okay kids you get ONE WEEK of Fable and ONLY AT 50%

Lazy Codex prompt to make it work harder: "Create an orchestration plan with a strong set of critical subagents (xhigh reasoning). You are done when all subagents report no unresolved concerns and have no critical comments."

@mattyglesias You can ask again after the final.

Where’s Haaland on the all-time ranking of Norwegians? I think he’s surpassed Amundsen but still behind Ibsen.

@RemyLevin @huseynovecon Haaland is always a beast

@huseynovecon I could see it happening. Not impressed with Brazil this year, and Haaland is a beast.

🇳🇴 (i'm believer!)

@thsottiaux It can't plan *as well* as ChatGPT 5.5 Pro Extended and that should be an option

What is something that you feel is surprising that Codex still can't do well and we should have gotten right a while ago?