My very first podcast...luckily I had great, thoughtful and eloquent guests in Temina Madon and @hanshengchia.bsky.social!

Temina, Markus, and I talk AI and development on this new podcast CGD

"We need to evaluate the specific behaviors our AI applications are designed to exhibit...then we can make claims about which actually led to better outcomes." @hanshengchia.bsky.social emphasizes the need for precise, ongoing evaluation of AI tools. Full remarks: https://bit.ly/42EF4VD

Alongside this year's #AnnualMeetings, experts are convening at CGD to unpack novel AI approaches and why they matter for policymakers. With: @hanshengchia.bsky.social, Rebecca Sharp, Temina Madon, Gabriel Demombynes, and @markusgold.bsky.social.#AnnualMeetings Happening Oct. 14—join us: https://bit.ly/42m2Vcl

There’s no consensus yet on what “good evaluation” means for AI-enabled interventions. To help raise the quality of evaluations, @hanshengchia.bsky.social & fellow experts propose 4 key investments, from setting standards to funding research to push the frontier: https://bit.ly/4n1Uxat

It's his imagination and we're just lucky to be living in it.

SecState Rubio claims that no on has died because of USAID cuts and suggested that I had lied in my reporting. So here I show him photos of specific children who have died because of the Trump administration's reckless cancelation of aid: www.nytimes.com/2025/05/31/o...

📅 On June 12, join @hanshengchia.bsky.social, Samantha Carter, & Farhan Abrol for a @voxdev.bsky.social event exploring the potential applications of AI. W/ @cgdev.org, @j-pal.bsky.social & @agencyfund.bsky.social. Register! ⬇️ cepr-org.zoom.us/webinar/regi...

AI-powered development interventions are growing in many areas, but underused. @hanshengchia.bsky.social, Surbhi Bharadwaj, & Christine Hwang urge governments to explore the potential of AI to transform e-portals & improve citizen access to digital public services: www.cgdev.org/blog/cutting...

AI is advancing fast, but evidence on what works for dev. is still emerging. @hanshengchia.bsky.social & @markusgold.bsky.social propose a 3-part framework to guide smart AI investments: 1️⃣ Supercharge proven solutions 2️⃣ Tackle known problems 3️⃣ Explore unknowns More: www.cgdev.org/blog/cutting...

Wrote this with @markusgold.bsky.social — we think there are 3 approaches that can guide smart investments in AI applications for development

"there's a lot of dumb AI-focused bullshit being pushed on users" and "this technology is not fake and has real use cases" are perfectly compatible things to believe

I’m often asked how AI can improve literacy or health if it makes mistakes/ behaves inconsistently. Evaluations help—but the term "eval" is vague. We outline a 4-stage framework to clarify. Feedback requested though also think many parts are practical and ready for adoption. 🙏 for collab J-PAL, TAF

The 2025 IMF & World Bank #SpringMeetings are right around the corner!🌸🌍 @cgdev.org is hosting key conversations with leading global development experts and policymakers—explore our schedule of events and sign up below ⬇️ bit.ly/43HcvIn

🥩

Sequel: Authoritarianism with American Characteristics

Thanks for inclusion in your roundup @thegovlab.org !

“Imagine the chief executives of America’s top universities and richest corporations…issued a public statement that spoke clearly and loudly about the dangers…” Those who speak out will at least put themselves on the right side of history. www.project-syndicate.org/commentary/w...

Meeting w/ nonprofits at the forefront of AI use, @hanshengchia.bsky.social finds that orgs are: 🌐Delivering services at scale with current AI tech 🌐Betting AI-driven personalization can boost impact 🌐Able to measure/manage AI unpredictability More:

The White House stated that USAID’s elimination would cause no deaths. A top official said (verbatim), “I give zero f*cks” about widespread credible reports of deaths already occurring. That official is incompetent. He did no analysis whatsoever of the death toll. Here is a serious analysis:

Just in India with orgs on frontier of nonprofit AI use. 3 insights: 1) deployments already happening at scale 2) orgs are betting on greater personalization driving impact 3) mitigations & evals for aberrant AI behavior exist but need to be demanded by funders. My blog for more: shorturl.at/6YKNx

This is one of those “how will this be reported if this was in another country” moments. Maybe another bilat aid agency should run a democracy, rights, and governance program for the US.

I will never be part of a more dramatic looking panel. Hong Kong cop show vibes

Rubio terminated 5800 USAID contracts – more than 90% of its foreign aid programs – in defiance of the courts. Here’s a list of just some of the lifesaving awards that were terminated. Nearly all were Congressionally mandated. They’ve saved millions of lives. 🧵

Per Dr Atul Gawande, ex USAID: “All malaria supplies protecting 53 million people, mostly children, including bed nets, diagnostics, preventive drugs, and treatments – terminated.” There are no words that sufficiently describe this level of cruelty and sadism.

I worked for Dean at USAID. Our office helped choose effective programs to get more bang for buck. This administration isn’t trying to improve spending, it’s “psychological warfare against a workforce that has been committed to furthering the lives of other people.” Here’s why he resigned.

FEWS NET is a really valuable tool for predicting famines before they happen. It is one of the reasons that famine is now no longer an act of God, but an act of war. Shutting it down sends a signal.

It’s crucial to develop a better understanding of the risks & opportunities AI presents for development. That's why CGD & Open AI are collaborators on the new AI for #GlobalDev Accelerator, headed by the Agency Fund. More from @hanshengchia.bsky.social & Dany Bahar: www.cgdev.org/blog/new-acc...

Amidst turmoil in foreign aid, we have to keep innovating, studying, and delivering services. Glad to be collaborating with The Agency Fund and OpenAI on a new accelerator for cost-effective AI applications. See Dany Bahar & my blog for more: www.cgdev.org/blog/new-acc...

As a non engineer, I’ve found the SWEs I’ve most enjoyed working with are high on “surprise factor”. Always new workarounds, new solutions, more elegant, more robust. I think driven by high curiosity, desire to understand micro problem + macro goal, and neatness. V fun collaborating with them

V fun discussing with the team behind UNDP’s human development report how we can get the upside from AI in development outcomes. Lots of potential for AI to impact things like literacy, nutrition, poverty but not simply b/c models getting more powerful. Will be doing lots more at @cgdev.org on topic

International migrants are the largest source of international aid. This is an often underappreciated benefit of migration for global poverty reduction.

Think a lot about causality with a baby. Did she fall asleep because of my superior soothing technique or b/c spouse tried 5 times before and tired her out? Sometimes ppl ask how being a new parent has been & I don’t know how to tell them it‘s lovely but with lots of micro, unsatisfying inferences

In 2 years would love to see cash+ where “+” are low cost AI agents users can access on demand with feature phone. E.g. neonatal coach advertised to moms who call after getting $; phone tutor; biz coach. Cash as base later relieves $ constraint, all “+” are low cost & drive agency,learning,referrals

Norway gives more foreign aid per capita than any other OECD country

Development programs aren’t built to adapt to the unique, varied needs of communities. @hanshengchia.bsky.social asks—can generative #AI help? 🌐 Read on ⬇ www.cgdev.org/blog/cutting...

Well, I believe the thing to say is “some personal news”. In January I will be leaving @cgdev.org after four and a half very happy years to take up the role of Deputy Chief Economist (Development) for the Foreign, Commonwealth and Development Office.

10 key insights from development economics research on @voxdev.bsky.social in 2024. Today I wrote about the standout takeaways from the broad range of research I have read this year. You can read the full blog on VoxDev: voxdev.org/topic/ten-ke... Here they are in a thread 🧵 1/11

Spoke at AE conference in Boston this week. I think AI for good discourse has been dominated by need to make AI safe, accurate, well governed & accessible. All v important, but missing key point about whether AI use translates into thing we care about (poverty, malnutrition reduction, literacy etc)

@hanshengchia.bsky.social & Alex Nawar have a few takeaways on the state of AI for development. They think that generative AI can: 🌐 Dramatically & affordably scale programs 🌐 Save workers time, allowing for more human interaction 🌐 Help important info flow up to authorities More ⬇ bit.ly/3Cid8fF

Greetings Bluesky. My new @cgdev.org post on potential of flipping the traditional aid model of big needs assessment >>> multi year development project to something a lot more adaptive, with gen AI