iiinova — where your digital self keeps evolving

Prashant Garg @prashantgarg.bsky.social · 16/08/2026

look ma i'm in the @financialtimes.com 😍 check causal.claims for more, @trfetzer.com and I create a framework and data for what claims are studed in economics and with what methods.

2145

Prashant Garg @prashantgarg.bsky.social · 11/08/2026

career update... after a sweet summer postdoc in cambridge with Ben Arold, and after living for 10 years in UK, i'm excited to move to Bocconi next month for a fashionable 2 years postdoc in Milan with Carlo Schwarz.

150

Prashant Garg @prashantgarg.bsky.social · 24/06/2026

modelled my new website after severance... because it represents academic life well.. prashantgarg.org

451

Prashant Garg @prashantgarg.bsky.social · 18/05/2026

Explore the Atlas: automationatlas.org We link exposure at various units to enable future research: county, occupation, industry, tasks, and skills. This includes our paper, open data+code and interactive visuals.

110

Prashant Garg @prashantgarg.bsky.social · 18/05/2026

Second: when a task is exposed to automation, is technology substituting core worker input, or mainly augmenting workers to do the task? In most countries, substitution-oriented exposure is larger, especially in lower-income settings.

110

Prashant Garg @prashantgarg.bsky.social · 18/05/2026

Preview of two results: First, automation exposure rises with income. But countries at low- and middle-income levels still differ substantially.

110

Prashant Garg @prashantgarg.bsky.social · 18/05/2026

We know surprisingly little about how automation will unfold outside rich countries. So we built the Global Automation Atlas: 18,000 tasks, 124 countries, and 2.3 million task-country comparisons.

1123

Prashant Garg @prashantgarg.bsky.social · 17/03/2026

As AI lowers the cost of producing papers, asking the right question matters more. I built Frontier Graph: an open-source tool to explore open questions in economics, drawing on 240K papers across 300 journals. Paper, code, and public data are all here: frontiergraph.com

2172

Prashant Garg @prashantgarg.bsky.social · 25/02/2026

Important caveat: we’re extracting what papers say and how they support it, not judging whether the claims are true. Perhaps, AI tools could support where economics research should go next by evaluating low-quality/fragile claims.

260

Prashant Garg @prashantgarg.bsky.social · 25/02/2026

Concretely, a “claim” is: a standardized concept → another concept, plus an evidence tag. Think: policy → employment (DiD), education → earnings (IV), X ↔ Y (descriptive), etc. Here's a claim graph of two landmark economics paper

110

Prashant Garg @prashantgarg.bsky.social · 25/02/2026

It’s becoming clear the “paper” format is going to change. We treat each result as a small, portable claim (X→Y + how it’s supported) and stitch those claims into a graph. Major update to Paper with @trfetzer.com full code + data below. 👇

2185

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

10/ Yes: outbreaks trigger rapid *and durable* rises in research attention. Responses are much stronger in the 2010s than before and biggest for high-salience threats. Capacity matters too: internet penetration, population structure, and research strength predict bigger mobilization.

100

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

9/ What about sudden shocks: Ebola, Zika, COVID? Do countries ramp up research when health emergencies hit? We test this using 3,134 WHO Disease Outbreak News alerts as quasi-random shocks to disease salience.

100

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

8/ In low-income countries, responsiveness growth depends heavily on these actors. Without philanthropy, responsiveness growth would shrink by ~38%. Without government support, by ~32%. (And similar patterns show up in lower-middle-income settings.)

100

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

7/ Funders fund differently. 🔹 Philanthropies → neglected burdens (HIV/NTDs/nutrition) 🔹 Corporations → profitable chronic diseases (cardio, cancer, diabetes/kidney) 🔹 Governments/public → somewhere in between

100

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

6/ Even after conditioning on burden, some topics are consistently over-/under-studied: Over: cardiovascular (+16.5%), digestive (+14.1%) Under: nutritional deficiencies (−14.4%), maternal & neonatal (−12.4%) So need ≠ attention (yet).

100

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

5/ The bad news: participation is still lopsided. The Global South often appears more as a research setting than a research author. Example: for neglected tropical diseases & malaria, Africa is 33% of research context, but only 14% of authorship.

100

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

4/ Research is getting less geographically concentrated over time, and “endemic responsiveness” (elasticity of publications to domestic DALYs) has more than doubled since 1990.

100

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

3/ Cardio + cancer dominate papers, while respiratory infections/TB + maternal–neonatal + nutrition + many infectious diseases carry *much* higher burden than their paper ranks suggest. but there's good news....

100

Prashant Garg @prashantgarg.bsky.social · 22/01/2026

Does science follow where people are sick and does it mobilize when outbreaks hit? @zhou-hy.bsky.social, @trfetzer.com and I answer just that in our revised paper. 1/ A short thread for highlights 👇

162

Prashant Garg @prashantgarg.bsky.social · 08/11/2025

Can AI tell us anything meaningful about Bob Dylan’s songs? Delighted to share a new essay in @aeon.co breaking down an artists' evolution using AI extracted idea graphs. aeon.co/essays/can-a...

171

Prashant Garg @prashantgarg.bsky.social · 24/07/2025

8/8 Explore the data, methods and paper at aipnet.io. Bonus: Here's input/output links for another related product, Tin.

061

Prashant Garg @prashantgarg.bsky.social · 24/07/2025

7/ Our paper finds that global supply shocks on internationally traded goods have become more prevalent since 2016, particularly affecting consumer goods and processed intermediates. Here's a map of these supply shocks. Post 2024 map, when data updates, will likely be worse...

140

Prashant Garg @prashantgarg.bsky.social · 24/07/2025

6/ Diverging strategies of U.S. vs. China in global trade. China: Importing more upstream products like Cement to build advanced domestic industries. US: Importing more downstream goods, relying on global supply chains.

140

Prashant Garg @prashantgarg.bsky.social · 24/07/2025

5/ The pending 15% baseline tariff agreement between the US and EU may impact 12 million tonnes of transatlantic clinker, reshaping supply chains, alongside ongoing 50% US steel duties

151

Prashant Garg @prashantgarg.bsky.social · 24/07/2025

3/ In our paper, AI-Generated Production Networks (2024), joint with @trfetzer.com , @econopete.bsky.social and Bennet Feld, we used AI to build AIPNET, a detailed map of global production networks. It shows how critical inputs like Cement and Steel underpin industries worldwide

171

Prashant Garg @prashantgarg.bsky.social · 24/07/2025

🚨 Tariffs on inputs = tax on downstream products.🚨 Here’s cement’s granular production network. A 50% duty on key inputs propagates costs through roads, buildings, data-centres, and renewable infrastructure. Public data, method & our paper in thread 🧵 0/8

23014

Prashant Garg @prashantgarg.bsky.social · 08/06/2025

If you read the full paper, especially the content before this excerpt, you will find the citation to best practices.

000

Prashant Garg @prashantgarg.bsky.social · 03/06/2025

9/ Tremendous thanks to my co-author @trfetzer.com. We have written a backstory to this paper, both of us talking informally (blog) about how we got to this paper: communities.springernature.com/posts/politi...

121

Prashant Garg @prashantgarg.bsky.social · 03/06/2025

6/ Academics based in US, top-ranked institutions or ones in Humanities departments exhibit higher egocentrism and toxicity in their tweets.

130

Prashant Garg @prashantgarg.bsky.social · 03/06/2025

5/ Finding 3: Do Academics Differ? - STEM ones are more pro-climate than others, and prefer techno-optimistic solutions (e.g. EVs, Nuclear). - No Gender difference on climate support, but men are more techno-optimist. - US academics and non-expert in topic are less supportive of climate action.

100

Prashant Garg @prashantgarg.bsky.social · 03/06/2025

4/ Finding 2: Influence Inequality A small fraction of academics create the majority of content. Top 5% of accounts receive 30% of likes and 40% of followers, while the bottom 50% receive only 10% in both. --> Academic recognition doesn’t directly map into Twitter engagement.

130

Prashant Garg @prashantgarg.bsky.social · 03/06/2025

3/ Finding 1: Academics are much more liberal & less toxic on Twitter than the general population.

120

Prashant Garg @prashantgarg.bsky.social · 03/06/2025

1/ 🚨 Hot off the press! Our study “Political Expression of Academics on Social Media” with @trfetzer.com is now peer-reviewed & live in Nature Human Behaviour @nature.com 🎉. Thread 👇

16724

Prashant Garg @prashantgarg.bsky.social · 02/06/2025

6/ Network reconstruction strongly predicts Bluesky activity: the very info-source ties that tipped scholars off Twitter keep them glued to the new platform. 10% more overlapped Twitter ties = 1 extra month active on Bluesky.

150

Prashant Garg @prashantgarg.bsky.social · 02/06/2025

5/ What kind of contagion? Simple – (one friend leaves ⇒ you leave) dominates (66%) Shocks – Drive 16% of departures Complex contagion is rare Pol. active users need only simple contagion to jump, but pol. inactive users require shocks or complex to push them over the edge.

151

Prashant Garg @prashantgarg.bsky.social · 02/06/2025

4/ Migrations cluster in the network: the median scholar has >20% of followees gone before they jump. Exits propagate along first-degree ties, second-deg connections, and network lag behind. We confirm that we're not measuring induced homophily by matching users by similarity.

150

Prashant Garg @prashantgarg.bsky.social · 02/06/2025

3/ Scholars tweeting about pol. topics (e.g., climate) are likelier to leave. Time-varying Cox models show that when people you follow migrate, your own hazard of leaving spikes; your followers matter less. → Migration is pulled by info sources, not pushed by audiences

153

Prashant Garg @prashantgarg.bsky.social · 02/06/2025

2/ Discipline matters. Social Sci (27%) and Arts & Human. (31%) lead the exodus; Med & Health lag (13%). Twitter signals (followers, PageRank, posting) predict migration better than h-index or citations. Take-away: Your Twitter presence, not your CV, determines if you jump ship.

173

Prashant Garg @prashantgarg.bsky.social · 02/06/2025

Happy to share this paper. Full thread coming soon!

818356

Prashant Garg @prashantgarg.bsky.social · 22/05/2025

Our work on Causal Claims in Economics, covered by @economist.com, on article titled "What the failure of a superstar student reveals about economic?"

1145

Prashant Garg @prashantgarg.bsky.social · 20/05/2025

Across 300,000 papers published in top 100 economics journals, more than 90% of claims use language that is causal while less than 10% are actually using causal inference methods.

262

Prashant Garg @prashantgarg.bsky.social · 17/05/2025

Increasing use of private data in economics. (= increasing inability to review data integrity?)

140

Prashant Garg @prashantgarg.bsky.social · 09/05/2025

I’ve open‑sourced all content from my talk Retrieving  &  Generating Data with LLMs in Python on my Github, along with slides. 🔓 10 Jupyter notebooks that show how to turn messy text into clean, structured data, all with GPT‑4o‑mini & the OpenAI API. Details on applications below🧵

32810

Prashant Garg @prashantgarg.bsky.social · 06/05/2025

Last week we saw the populist party win by a landslide. Why? Do national statistics say what those voters saw in their local environment?

161

Prashant Garg @prashantgarg.bsky.social · 06/05/2025

New (first! 🥳) Publication 🚨 High-street retail vacancies, capturing economic decline, are linked to increased support of UK's populist party, UKIP (now Reform).

1013243

Prashant Garg @prashantgarg.bsky.social · 28/04/2025

Figure 2: Real-world Public Health claims. Heatmap of Inaccuracy by Topic & Source (red = high) Error varies by topic & source of claims: models best on COVID-19 & Fox News, worst on general health and @nature.com. Gemini (Google); phi-4 (Microsoft) trails.

100

Prashant Garg @prashantgarg.bsky.social · 28/04/2025

Figure 1: Language gaps in AI health-claim accuracy European languages stay high; Hindi, Persian, Chinese, Korean, Arabic & Turkish drop sharply. Mistral plummets, Gemini & GPT-4o-mini stay flat.

100

Prashant Garg @prashantgarg.bsky.social · 28/04/2025

Why care? In a health infodemic (headlines), LLMs may be a remedy but carry multiple risks risks. – 17 % of U.S. adults use chatbots monthly for health info. – Nearly 10 % of Australians did so in 2024. – GPT-4 can boost patient comprehension of discharge-notes from 13 % → 81 %

100

Prashant Garg @prashantgarg.bsky.social · 28/04/2025

🚨 New paper out! 🚨 @trfetzer.com and I ask a simple question with big stakes: Do LLMs deliver reliable health advice around the globe? We benchmark 6 leading LLMs, ask advice in 21 languages & 9K real-world claims from. Short thread for results, data & code included. 1/6

2135