The computational psych preconference is back @spspnews.bsky.social for a full day! This year's lineup:
👉theory-driven modeling: Hyowon Gweon
👉data-driven discovery: @clemensstachl.bsky.social
👉application: me
👉 panel: @steveread.bsky.social Sandra Matz, @markthornton.bsky.social Wil Cunningham
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.
🚨 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
New Commentary in @misinforeview.bsky.social
We argue that recent technological disruptions provide an opportune moment for self-reflection, bringing together #AI, #science #misinformation, and #social #disparities within one research agenda. 🌎 🤖
misinforeview.hks.harvard.edu/article/disp...
And that’s a wrap on the day! Our workshop keeps growing, and so do our social events.
Huge thanks to all the authors, speakers, attendees, and of course, our amazing organisers 👏
The drinks were a blast!
#ACL2025NLP
The 9th edition of the Workshop for Online Abuse and Harms is kicking off in Room 2.17 at #ACL2025NLP! We have a full program today, including 3 invited speakers, a panel discussion and a poster session. This year's workshop theme is Harms Beyond Hate Speech.
Our paper on #ChildObjectification on TikTok is finally out! Check out my poster presentation tomorrow at the #Workshop on #Online #Abuse and #Harms @ #ACL.
⬇️ Read more about our findings in the post below!
aclanthology.org/2025.woah-1....
What an inspiring first day at the Summer School for Women* in Political Methodology here in Bremen yesterday! We were honored to be welcomed by the president of @unibremen.bsky.social, Prof. Jutta Günther, (1)
The #SocialMedia track is starting at #IC2S2 with a talk on data donations, followed by presentations on polarization, engagement patterns, and community rules! 🌐
Join our Centre for Sociology of Humans & Machines (SOHAM) at Trinity College Dublin as a Postdoctoral Research Fellow, exploring how AI can enhance collective intelligence. We're looking for someone with strong computational skills & a passion for human-machine systems
my.corehr.com/pls/trrecrui...
The paper will be out with the ACL proceedings soon, I’ll keep you posted !
A photo of Olstugan Tullen
I’ve never seen so many people at my institute‘s favorite bar — #ic2s2 is going strong day and night. Amazing people, amazing vibes!
Topic Overview of Comments on Videos with Children
Join my talk on #ChildObjectification on TikTok at #IC2S2 today at 2:30 PM (📍 Social Good & Ethics).
I built classification models to detect objectifying language and found: 10% of comments refer to appearance, 3% are objectifying. Models struggle with this task, with #RoBERTa outperforming GPT-4.
If you're @ic2s2.bsky.social #IC2S2 and are looking for a faculty position, get in touch. @uni-saarland.de is hiring.
Right now, there's an open call in Computer Science @saarland-informatics-campus.de: www.uni-saarland.de/fileadmin/up...
Soon, there'll be open calls in the Social Sciences.
We just kicked off our tutorial at #IC2S2 in beautiful Norrköping. @dustinbwright.com is currently presenting our hands-on part on automated fact checking - 📍stop by at Vingen 3-4!
I had a great time at GSO's #LeadershipAcademy in Berlin! Leading other people is not always straightforward, especially in academia.
My takeaway: merging #values around who I want to be as a mentor with a strategy for a lab in #ComputationalViolenceResearch I hope to build.
Thanks for inviting me and for the mug! ☕️
Being a good researcher goes beyond academic publishing; to me, it also means fostering a research environment grounded in integrity and continuous learning. That’s why I’m excited to be part of this year’s GSO #Leadership Academy. Looking forward to meeting this year’s cohort! #ResponsibleResearch
Have you ever asked yourself about the overall extent of TikTok? Here some numbers from "Just Another Hour on TikTok" - Great compliment to @bendavidsteel.bsky.social for this data collection effort!
w/ @miriamschirmer.bsky.social & Derek Ruths
arxiv.org/abs/2504.13279
Join us online for my #TaDa talk on child exposure on TikTok next week on Wednesday! 🤓
Job offer! 3-year PDRA studying archaeological content on social media from a cultural evolution perspective. Based at Edinburgh, supervised by @chiara-bonacchi.bsky.social and myself -project "Weaponised Pasts: The Evolution Of Heritage-based Hostility On Social Media’. Please share!
bit.ly/3QCXUFS
🤝 Human-AI synergy remains essential: AI speeds up analysis and identifies overarching themes. Humans provide interpretative depth, socio-political and historical awareness, and ideological sensitivity crucial in discourse analysis.
💡 Our Suggestion: To maximize the potential of LLMs, combine AI’s scalability with human expertise through culturally-informed training and theory-driven prompting.
❗️ The Problem: AI neutrality often obscures deeper socio-political meanings. The model often misclassifies ideologically charged statements, underplaying their socio-political implications.
🔍 Key Findings:
GPT-4 efficiently identifies broad themes in large datasets but struggles with ideologically charged and culturally sensitive discourse. The model tends to default to neutral categories, failing to explicitly recognize hate speech or exclusionary narratives.
🧵New paper out! How effective is GPT-4 for analyzing hate speech on social media? Our research in the Journal of Multicultural Discourses explores this important question. www.tandfonline.com/doi/full/10....
Great to be part of this spring's #TaDa Speaker Series - looking forward to the talks this season!
📕 My dissertation on #NLP for #Violence Studies has been published: mediatum.ub.tum.de?id=1751256
I've been looking at #abusive behavior online, as well as sharing of personal experiences with violence, incl. psychological #trauma.
Excited to push this research forward and connect with others 🌐
JOIN 2025 MethodsNET Summer School with Central European University (CEU)
📍 Central European University, Vienna
🗓️ 30 June to 18 July 2025
The full course list is online!!
More details: bit.ly/4gLbv9V
Early Bird Registration: Early February, 2025
#research #researchmethods #methodology #methods
Job announcement: Junior Professorship in Computational Social Science at the University of Konstanz.
Join us at the Center for Data and Methods! Feel free to reach out if you have questions.
stellen.uni-konstanz.de/jobposting/2...
We are hosting the 11th International Conference on Computational Social Science in Sweden
🚀The IC2S2'25 website is LIVE, and submissions are OPEN!
📍Norrköping | July 21-24, 2025
Call for Abstracts (until Feb 24)
Call for Tutorials (until Jan 17)
🔗Explore details & submit: ic2s2-2025.org
