Laurence Dierickx

WSJ reporting that the U.S. used Claude for the air strikes in Iran. Centcom has been using Claude "for intelligence assessments, target identification and simulating battle scenarios" www.wsj.com/livecoverage...

NYU is hiring two postdocs at the Center for Social Media, AI & Politics:
csmapnyu.org/jobs

And a grant manager in Sociology with the Social Science Research Hub:
uscareers-nyu.icims.com/jobs/15327/g...

I'm part of both groups--please share with anyone who is interested!

About 12% of U.S. teens turn to AI for emotional support or advice | TechCrunch General purpose tools like ChatGPT, Claude, and Grok are not designed for this use, making mental health professionals wary.

General purpose tools like ChatGPT, Claude, and Grok are not designed for this use, making mental health professionals wary.

Thank you so much!

I am excited to announce that on 2 March I will begin a MSCA COFUND Fellowship at the University of Bergen as part of the LEAD AI programme in the Department of Information Science and Media Studies. My research project focuses on search engines as AI-mediated environments in journalism.

A small number of samples can poison LLMs of any size Anthropic research on data-poisoning attacks in large language models

A small number of samples can poison LLMs of any size www.anthropic.com/research/sma...

I was aware of the privacy issues on LinkedIn, but I didn't think it was this so serious and that the platform presented so many potential violations of the GDPR and the AI Act. thelocalstack.eu/posts/linked...

Ways of working with the Wayback Machine - Internet Archive Europe Guest post by Jonathan W. Y. Gray. How do researchers, journalists, and artists work with archived pages from the Internet Archive's Wayback Machine? What kinds of tools, methods, and approaches do th...

How do researchers, journalists & artists work with the Wayback Machine?
At a recent gathering in Amsterdam, we explored how web archives support digital investigations, platform research & media art — and how we can build better tools & collaborations.
👉 www.internetarchive.eu/2026/02/18/w...

What is a fact in the age of generative AI? Fact-checking as an epistemological lens Generative AI systems produce outputs that are coherent and contextually plausible yet not necessarily anchored in empirical evidence or ground truth. This challenges traditional notions of factual...

(6) As such, our paper introduces a new theoretical framework that advances debates in epistemology, AI ethics/literacy, and media studies. Enjoy! www.tandfonline.com/doi/full/10....

(5) We conceptualized these as emergent facts, arising from the interaction of training data, model architecture, and user prompts.

(4) interpretative facts, which are derived from context, analysis or reasoning; and rule-based facts, which are defined by formal systems, logic or procedural rules. While each type of fact provides valuable insight, AI-generated content do not fit neatly into these categories.

(3) Upon reflecting on the epistemology of fact-checking, we found that traditional fact-checkers deal with three types of facts: evidence-based facts, which are grounded in empirical data and observable reality...

(2) Although generative AI systems produce coherent and contextually plausible outputs, these are not necessarily grounded in empirical evidence or 'truth'. This calls into question traditional ideas about what constitutes a fact in computational contexts. That was our starting point.

🚨Publication alert! (1) I am delighted to present our latest study with my amazing colleagues from the University of Bergen: What is a fact in the age of generative AI? Fact-checking as an epistemological lens. 🧵

This paper was conceived and developed at the University of Bergen within the NORDIS Hub @nordishub.bsky.social - research was funded by EU CEF grant number 10115860 cc @edmo-eu.bsky.social

What is a fact in the age of generative AI? Fact-checking as an epistemological lens Generative AI systems produce outputs that are coherent and contextually plausible yet not necessarily anchored in empirical evidence or ground truth. This challenges traditional notions of factual...

(6) As such, our paper introduces a new theoretical framework that advances debates in epistemology, AI ethics/literacy, and media studies. Enjoy! www.tandfonline.com/doi/full/10....

(5) We conceptualized these as emergent facts, arising from the interaction of training data, model architecture, and user prompts.

(4) interpretative facts, which are derived from context, analysis or reasoning; and rule-based facts, which are defined by formal systems, logic or procedural rules. While each type of fact provides valuable insight, AI-generated content do not fit neatly into these categories.

(3) Upon reflecting on the epistemology of fact-checking, we found that traditional fact-checkers deal with three types of facts: evidence-based facts, which are grounded in empirical data and observable reality...

(2) Although generative AI systems produce coherent and contextually plausible outputs, these are not necessarily grounded in empirical evidence or 'truth'. This calls into question traditional ideas about what constitutes a fact in computational contexts. That was our starting point.

🚨Publication alert! (1) I am delighted to present our latest study with my amazing colleagues from the University of Bergen: What is a fact in the age of generative AI? Fact-checking as an epistemological lens. 🧵

OpenAI retired its most seductive chatbot – leaving users angry and grieving: ‘I can’t live like this’ Its human partners said the flirty, quirky GPT-4o was the perfect companion – on the eve of Valentine’s Day, it’s being turned off for good. How will users cope?

People grieving AI companions are not confused about reality. Attachment forms through interaction, memory, and emotional rhythm. When companies design for intimacy but control continuity, shutdown becomes relational harm, not just product change. 1/2 www.theguardian.com/lifeandstyle...

AI as Cultural Mediators: Rethinking Environmental Intelligence and Human-Machine Interdependence

AI as Cultural Mediators: Rethinking Environmental Intelligence and Human-Machine Interdependence hdsr.mitpress.mit.edu/pub/mmunexaf...

Ben was the lead data journalism developer at the LATimes for many years. He’s open sourcing his new news apps course materials!

Oui, c'est effrayant et ça l'est d'autant plus que ce rédacteur en chef intervient régulièrement dans certaines écoles de journalisme. Comme quoi la route vers l'éducation à l'IA est encore longue et semée d'embûches !

Cette déclaration suggère l’une de ces deux choses :
– une incompréhension du fonctionnement des LLMs, ou
– une inquiétante complaisance à l’égard d’une génération probabiliste traitée comme un jugement éditorial.
Le même trouve que les "facultés" de l'IA sont "stupéfiantes". Sans voix. (2/2)

En Belgique, le rédacteur en chef de ce qui est considéré un quotidien de qualité reconnaît, en toute transparence, demander à l'IA générative "ce qu'il pense d'un projet de texte. Je le soumets à l'IA comme je le soumettrais à un journaliste de la rédaction." (1/2)

1) a misunderstanding of how large language models actually work, or 2) a troubling comfort with probabilistic text generation being treated as editorial judgment. Either way, that’s pretty scary. (2/2)

In Fr. Belgium, the editor-in-chief of a newspaper widely regarded as a quality publication admits using generative AI to ask: “What it thinks of a draft text. I submit it to AI the same way I would submit it to a journalist in the newsroom.” This statement suggests one of two things: (1/2)

Claims of “intelligence” focus on function: if AI behaves “as if” it has a mind, it’s “intelligent.” But behavior is useless when models are designed to pass behavioral tests. Instead, the AGI label achieves the opposite goal: it suggests a human-like experience of being an LLM. ⬇️