Lightnews — Scholar-powered news

Reposted by Stefano Sangiovanni

Nicolai Berk @nicolaiberk.bsky.social · 9d

I had the pleasure to teach a 3-day crash course on #LLMs for PhD students.

We covered:

1️⃣ Text Representation and Embeddings
2️⃣ Machine Learning & Transformer Architecture
3️⃣ Generative Models for Social Sciences

The 6 slide sets and 10 notebooks are available on github: github.com/nicolaiberk/...

GitHub - nicolaiberk/llm_ws: Materials for my Workshop on LLMs

Materials for my Workshop on LLMs. Contribute to nicolaiberk/llm_ws development by creating an account on GitHub.

github.com

3 9 63

Stefano Sangiovanni @stesangio.bsky.social · 14d

Had the blessing to spend last spring at MZES as a visiting PhD student and it was a BLAST!

Brilliant scholars, a super supportive environment, and the perfect place to be productive (also have fun!) 🌈

MZES Uni Mannheim @mzesunimannheim.bsky.social · 16d

📣 Attention, postdoctoral researchers!

❗ Apply now for our MZES Visiting Fellowships 😊

💡 Spend 2-4 weeks at the MZES to share ideas
💰 Funding for accommodation, travel, daily allowance
📆 Deadline: 28 November

Full information:
👉 www.mzes.uni-mannheim.de/en/news/deta...

Screenshot of the call for applications for MZES Visiting Fellowships, 22 September 2025. For the full text, please follow the link.

1 6

Reposted by Stefano Sangiovanni

Stefano Sangiovanni @stesangio.bsky.social · 19d

Checked a random one just for fun during my lunch break, same dataset: dmsjournal.biomedcentral.com/articles/10....

In Supplement 2, sdLDL-c→WC (T2D) is β=0.008 with CI (−0.12, −0.14) 🤔🤔🤔
In Table 3 (WC), lbLDL-c is β=0.03 with CI (−0.03, 0.03) strange CIs?? 🤔

Thanks for your service Sophie!

1 1 1

Stefano Sangiovanni @stesangio.bsky.social · 19d

Sounds like a great plan! Beyond the formatting typos, an LLM can instantly spot odd CIs or effects outside of them. I literally opened a random paper.. If I speak im in trouble 🥶

1

Stefano Sangiovanni @stesangio.bsky.social · 19d

Checked a random one just for fun during my lunch break, same dataset: dmsjournal.biomedcentral.com/articles/10....

In Supplement 2, sdLDL-c→WC (T2D) is β=0.008 with CI (−0.12, −0.14) 🤔🤔🤔
In Table 3 (WC), lbLDL-c is β=0.03 with CI (−0.03, 0.03) strange CIs?? 🤔

Thanks for your service Sophie!

1 1 1

Reposted by Stefano Sangiovanni

Sophie E. Hill @sophieehill.bsky.social · 19d

I regret to inform you I found *another* paper by (some of) these authors with similar errors...

🧵

Sophie E. Hill @sophieehill.bsky.social · Sep 8

⛏️ After a LOT of digging, here's what I found:

🚩 5 estimates outside their CI
🚩 46 estimate-CI duplicates
🚩 20+ cases of asymmetric CIs
🚩 17 values rounded to 4 decimal places (all others 3dp)
🚩 >50% of all estimates and confidence intervals are multiples of 0.008 (???)

(CI = confidence interval)

2 9 46

Reposted by Stefano Sangiovanni

Joachim Baumann @joachimbaumann.bsky.social · 26d

🚨 New paper alert 🚨 Using LLMs as data annotators, you can produce any scientific result you want. We call this **LLM Hacking**.

Paper: arxiv.org/pdf/2509.08825

$We present our new preprint titled "Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation". We quantify LLM hacking risk through systematic replication of 37 diverse computational social science annotation tasks. For these tasks, we use a combined set of 2,361 realistic hypotheses that researchers might test using these annotations. Then, we collect 13 million LLM annotations across plausible LLM configurations. These annotations feed into 1.4 million regressions testing the hypotheses. For a hypothesis with no true effect (ground truth $p > 0.05$), different LLM configurations yield conflicting conclusions. Checkmarks indicate correct statistical conclusions matching ground truth; crosses indicate LLM hacking -- incorrect conclusions due to annotation errors. Across all experiments, LLM hacking occurs in 31-50\% of cases even with highly capable models. Since minor configuration changes can flip scientific conclusions, from correct to incorrect, LLM hacking can be exploited to present anything as statistically significant.$

5 94 260

Reposted by Stefano Sangiovanni

Sophie E. Hill @sophieehill.bsky.social · Sep 5

"Sweeteners can harm cognitive health equivalent to 1.6 years of ageing, study finds"

Or does it? Let's take a look at this "study"...

www.theguardian.com/food/2025/se...

Sweeteners can harm cognitive health equivalent to 1.6 years of ageing, study finds

Researchers say low- and no-calorie sweeteners appear to affect thinking and memory in middle age

www.theguardian.com

25 170 470

Stefano Sangiovanni @stesangio.bsky.social · Sep 6

Had a great time at #SISP25 in Naples! :D

Presented a couple of papers on valence politics and political scandals, plus a bonus talk on AI & audio-as-data in the Research Methods workshop!

It was wonderful to catch up with so many friends 💜

Giorgio Dolci @giorgiodolci.bsky.social · Sep 6

Quality time with the NASP-UniMi crew in Naples #SISP25

@stesangio.bsky.social

3

Reposted by Stefano Sangiovanni

Enrico Calossi @enricocalossi.bsky.social · Sep 5

Stefano Sangiovanni presents the paper "Party Competition and the Strategic Use of Valence Signals in Electoral Campaigns" in the panel on party organization.
#Sisp2025 #NapleSisp @stesangio.bsky.social

1 2

Reposted by Stefano Sangiovanni

Julius Diener @dienerjulius.bsky.social · Aug 25

Have you ever wondered if young politicians perceive their roles as representatives differently than their older colleagues? Then, my new research note in Party Politics may interest you. I investigate age differences in the representation styles of politicians.
journals.sagepub.com/doi/10.1177/...

Sage Journals: Discover world-class research

Subscription and open access journals from Sage, the world's leading independent academic publisher.

journals.sagepub.com

4 8 41

Reposted by Stefano Sangiovanni

John Holbein @johnholbein1.bsky.social · Aug 4

Assistant professors teach courses that are closer to the knowledge frontier than tenured faculty.

2 9 35

Reposted by Stefano Sangiovanni

Terence Tao @teorth.bsky.social · Jul 19

My thoughts on the crucial importance of methodology on self-reported AI performance on mathematics competitions, and my policy on commenting on such reports going forward: mathstodon.xyz/@tao/1148814...

Terence Tao (@[email protected])

It is tempting to view the capability of current AI technology as a singular quantity: either a given task X is within the ability of current tools, or it is not. However, there is in fact a very wid...

mathstodon.xyz

2 52 230

Reposted by Stefano Sangiovanni

Fabrizio Coticchia @fabrizio-coticchia.bsky.social · Jul 23

Dichiarazione della Società Italiana di Scienza Politica (SISP) su Gaza

2 6

Reposted by Stefano Sangiovanni

Giorgio Dolci @giorgiodolci.bsky.social · Jul 23

Super happy to share my new (well, actually first ever!) publication!
In this article, @giacomomelli.bsky.social and I set out to operationalize and validate a scale of populist attitudes, using data from Round 10 of the #EuropeanSocialSurvey.

Full article here:
link.springer.com/article/10.1...

Measuring populism in Europe. Comparison and validation in the European Social Survey - Quality & Quantity

Quality & Quantity - Populism has become a defining feature of European democracies, prompting extensive scholarly debate about its impact on political processes. This study refines the...

link.springer.com

2 10

Reposted by Stefano Sangiovanni

Kobi Hackenburg @kobihackenburg.bsky.social · Jul 21

Today (w/ @ox.ac.uk @stanford @MIT @LSE) we’re sharing the results of the largest AI persuasion experiments to date: 76k participants, 19  LLMs, 707 political issues.

We examine “levers” of AI persuasion: model scale, post-training, prompting, personalization, & more!

🧵:

10 56 110

Reposted by Stefano Sangiovanni

EPSS @epssnet.bsky.social · Jun 26

🚨 Big News for European Political Science 🚨

We’re thrilled to announce the launch of the European Political Science Society (EPSS): a new, member-led, not-for-profit association built to support our scholarly community.

🔗 epssnet.org

Here’s a thread with everything you need to know.

🧵

4 210 380

Stefano Sangiovanni @stesangio.bsky.social · Jun 18

Congratulazioni Konstantin, complimenti! 🎉

1 1

Reposted by Stefano Sangiovanni

Stefan Müller @stefanmueller.bsky.social · Jun 5

📄 @fgilardi.bsky.social created this template for writing abstracts several years ago, and I’ve tried to follow Fabrizio‘s suggestions ever since.

PDF: fabriziogilardi.org/resources/pa...

Everyone agrees that this issue is really important. But we do not know much about this specific question, although it matters a great deal, for these reasons. We approach the problem from this perspective. Our research design focuses on these cases and relies on these data, which we analyze using this method. Results show what we have learned about the question. They have these broader implications.

2 35 97

Reposted by Stefano Sangiovanni

Marius Sältzer @msaeltzer.bsky.social · Jun 5

Hey y'all I am hiring a PhD student. Minimal teaching load and many opportunities for more ;)

uol.de/job566en

Research Associate in the field of "Digital Social Science" // University of Oldenburg

uol.de

1 26 22

Stefano Sangiovanni @stesangio.bsky.social · May 29

MUST READ. Super super relatable.

I use the "Rant Technique" a lot with my partner and colleagues. It really helps us feel like we’re all in the same boat, not just on the academic journey, but in life in general.

Catherine E. de Vries @catherinedevries.bsky.social · May 29

My most personal post yet, a response to so many messages after launching Respect the Marble

“Breaking the Intellectual Doom Loop: How to Write While the World Burns”

ft. anxiety, a doom-containment protocol & heroic attempts to write offline

🧵

catherineeunicedevries.substack.com/p/breaking-t...

Breaking the Intellectual Doom Loop: How to Write While the World Burns (and Your Brain Spirals)

Featuring Anxiety, a Doom-Containment Protocol, and a Heroic Effort to Work Offline

catherineeunicedevries.substack.com

1 1 5

Reposted by Stefano Sangiovanni

Sofie Cairo @cairosofie.bsky.social · May 22

🚨 New WP! 📄 "Publish or Procreate: The Effect of Motherhood on Research Performance" (w/ @valentinatartari.bsky.social
👩‍🔬👨‍🔬 We investigate how parenthood affects scientific productivity and impact — and find that the impact is far from equal for mothers and fathers.

2 100 210

Reposted by Stefano Sangiovanni

Alessandro Nai @alessandronai.bsky.social · May 26

We are looking for a PhD student!

Check out the cfa below, please repost for N and diversity

👇👇🔥🔥🔥

APSA & ICA Political Communication Divisions @polcomm.bsky.social · May 26

PhD opportunity!

ASCoR is looking for a motivated PhD candidate for the project “Visual Politics and the Emergence of Angry Publics.” This four-year PhD track will be supervised by @ulrikeklinger.bsky.social, @lindabos.bsky.social, and @alessandronai.bsky.social DL: June 4

shorturl.at/YeXpf

Vacancy — PhD Position "Visual Politics and the Emergence of Angry Publics"

Are you curious about how images, videos, and memes transform political communication? Are you passionate about digital technology and the future of democracy? Then apply for this PhD project focused ...

werkenbij.uva.nl

19 22

Reposted by Stefano Sangiovanni

Flaminio Squazzoni @squazzoni.bsky.social · May 21

Do you want to learn how to apply #computational methods to study social and political #dynamics? We have a new Master's programme in #Milan. Join the presentation of this programme on 27 May 11:30 CET time by registering to the link below! cosp.orientamentounimi.it/en/dettagli_...

1 2

Reposted by Stefano Sangiovanni

MZES Social Science Data Lab @mzes-ssdl.bsky.social · May 21

🕑 Join us *today* at 13:45 CEST for the talk 'Forecasting the German Federal Election 2025 - Different Modelling Approaches' by @hannahrajski.bsky.social (University of Mannheim) & @cornelius-erfort.bsky.social (Witten/Herdecke University).

⬇️ Zoom link and details below

MZES Social Science Data Lab @mzes-ssdl.bsky.social · May 15

🚨 Upcoming: "Forecasting the German Federal Election 2025 - Different Modelling Approaches"

👤 @hannahrajski.bsky.social (Uni Mannheim) & @cornelius-erfort.bsky.social (Witten/Herdecke Uni)

🗓️ Wed, May 21, 13:45-15:15 CET

📺 Register for the live stream: us02web.zoom.us/meeting/regi...

Forecasting the German Federal Election 2025 - Different Modelling Approaches

Abstract

The Zweitstimme.org project presents a comprehensive approach to forecasting the 2025 German federal election, combining multiple methodologies to predict both party vote shares (Zweitstimme) and constituency-level outcomes (Erststimme). This presentation showcases three distinct forecasting approaches: a dynamic Zweitstimme model, a proportional swing Erststimme model, and citizen forecasting. The project also communicated the forecasts and their uncertainties to the public, with results being shared in various media outlets.

Presenter(s)

Hannah Rajski is a PhD candidate in Political Science at the Graduate School of Economic and Social Sciences at the University of Mannheim. She works in the project Election Forecasts for the German Federal Election 2025. Her main research interests lie in quantitative methods, comparative politics, and political sociology with a focus on political behaviour and elections. In her dissertation she analyses voters‘ expectations about election outcomes and how these can be aggregated into citizen forecasts. Hannah holds a bachelor’s degree in Politics, Administration and International Relations from Zeppelin University, Friedrichshafen, Germany and a master’s degree in Politics and Public Administration from the University of Konstanz, Germany.

Cornelius Erfort is Postdoc at the Witten/Herdecke University, working on the project Election Forecasts for the German Federal Election 2025. His general interests are in comparative politics and quantitative methods. More specifically, he is working on voter targeting, interest groups, and voting behavior. In his dissertation, he analyzed how parties use the target and tailor their digital election ads. He was a member of the Research Training Group DYNAMICS which is jointly organized by Humboldt-Universität zu Berlin and the Hertie School.

3 8