Sune Lehmann

So good!

IceLab Camp Part of the Stress Response Modeling Research School, IceLab Camp is a four-day off-site PhD course that prepares its participants to create new inter- or multidisciplinary research by first teaching ...

Registration is now open for IceLab Camp – a four-day off-site PhD course designed to train participants in asking research questions, laying the foundation for new multidisciplinary collaborations. www.umu.se/en/icelab/ca...

Image of an abstract with the following text Abstract: A year ago, ChatGPT surprised the world with its extraordinary language generation capabilities. Chatbots have since become of the fastest adopted consumer product in history with investments in genAI forecasted to reach $12B this year. In this talk, I will first review the fast-evolvingliterature on the document-level membership inference task for LLMs: the methods proposed to detect--a posteriori--whether a specific piece of text was seen during training by an LLM and at least partially memorized, the distribution shift concerns, and some of the solutions proposed. I will then discuss the use of randomized controlled setups to causally study LLM memorization. In particular, Iwill discuss how randomized controlled setup have shed lights on the determinant of memorization and showed LLMs to have a mosaic memory. I will conclude the talks with same thoughts on the security and privacy challenges ahead when it comes to LLMs and the use of synthetically generated trap sequences for membership inference.

Speaker: Yves-Alexandre de Montjoye
Time: 8 Apr. 2025, 15:00-16:00
Place: SODAS Conference Room - 1.1.12, Øster-Farimagsgade 5
Title: LLM's unintended memories

sodas.ku.dk/events/llms-...

People of Copenhagen:

On Tuesday April 8th, we have awesome privacy researcher @yvesalexandre.bsky.social visiting the group. Yves is a bold and creative scientist, and also former advisor to Marianne Vestager.

Yves will give a talk at SODAS at 3pm that's open to the public (details below)

Today is the (hard) deadline for the postdoctoral call listed below! Join us #postdoc #scienceofscience #AI #networkscience

Congratulations! Well deserved :)

Associate Professor or DTU Tenure Track Assistant Professor in AI - DTU Compute Are you passionate about Human Centered AI and can you help build the next generations of AI systems within the Danish Pioneer Centre for AI? DTU Compute has an attractive group leader opening as Tenu...

Job opening at DTU Compute & @aicentre.dk - Associate Professor or Tenure Track

Forskere: Vi står over for en teknologisk revolution, og Danmark kan vise vejen Med sine stærke traditioner for fællesskab og tillid kan Danmark vise, at ai både kan skabe værdi og gøre samfundet bedre.

New Politiken op-ed (in 🇩🇰) by @adler-nissen.bsky.social, @sunelehmann.com, Morten Axel Pedersen and other colleagues. They talk about how AI can be used democratically, innovatively, justly and with the interests of Danish citizens in mind. Read on: politiken.dk/debat/kronik...

🇩🇰 🤖 🌎 In yesterday's @politiken.dk we shared an op-ed outlining our vision for CAISA - Denmark's National Center for Artificial Intelligence in Society, led by @adler-nissen.bsky.social and deputy director Professor Thomas Moeslund (1/10)

Ph.D. and postdoc positions in Data Science for epidemic preparedness at the NERDS research group The NERDS (NEtwoRks, Data, and Society) team at the IT University of Copenhagen welcomes applications from aspiring PhD students and postdocs in the areas of Da

I am hiring PhD students and Postdocs to join me in beautiful Copenhagen. Together, we will develop data science methods to improve epidemic preparedness.

Copenhagen is amazing, salary is good, we have plenty of funding, and a great community.

Read more..: candidate.hr-manager.net/ApplicationI...

🚨 Cool job alert in #DataScience

If you're looking for a PhD or postdoc position then take a look here 👇

Jonas is a fantastic researcher, you will be embedded in a fantastic research group, and work on highly relevant and applicable research

🌍 Join us in Copenhagen - fantastic city for work-life balance and awesome science.

I am hiring PhD students & Postdocs in my group at SODAS Univ of Copenhagen to explore AI, network science & the science of science.

📍 Start: flex summer 2025
📌 Info&Apply: www.robertasinatra.com/2025/03/02/p...

Executive Summary A pro-Russia content aggregation network, Pravda, appears to be set up to flood large-language models with pro-Kremlin content, The American Sunlight Project has found. Over the past several months, ASP researchers have investigated 108 new domains and subdomains belonging to the Pravda network, a previously-established ecosystem of largely identical, automated web pages that previously targeted many countries in Europe as well as Africa and Asia with pro-Russia narratives about the war in Ukraine. ASP’s research, in combination with that of other organizations, brings the total number of associated domains and subdomains to 182. The network’s older targets largely consisted of states belonging to or aligned with the West. Notably, this latest expansion includes many countries in Africa, the Asia-Pacific, the Middle East, and North America. It also includes entities other than countries as targets, specifically non-sovereign nations, international organizations, audiences for specific languages, and prominent heads of state. The top objective of the network appears to be duplicating as much pro-Russia content as widely as possible. With one click, a single article could be autotranslated and autoshared with dozens of other sites that appear to target hundreds of millions of people worldwide. ASP researchers also believe the network may have been custom-built to flood large language models (LLMs) with pro-Russia content. The network is unfriendly to human users; sites within the network boast no search function, poor formatting, and unreliable scrolling, among other usability issues. This final finding poses foundational implications for the intersection of disinformation and artificial intelligence (AI), which threaten to turbocharge highly automated, global information operations in the future.

A pro-Russia content aggregation network is churning out at least 3 MILLION pieces of propaganda per year, all on sites that are virtually unusable by humans.

So what's the goal? We explore the idea that it might be to flood LLMs with pro-Russia content:
static1.squarespace.com/static/6612c... 1/

Visiting @mpidr.bsky.social this week—super excited to see what’s happening in Demographic Studies (don’t miss my talk!).
Also, I’ll be in Berlin on Feb 2, Helsinki from Feb 3-6, and Copenhagen from Feb 10-12. Let me know if you’re around and up for a coffee 🧪🔬☕️

Nice!!

How to make your own Die Hard Christmas tree ornament. We’ve all by now seen this excellent Die Hard Christmas tree ornament, but since it got spread around virally so late in the year no one had any time to sell them (ahem, Etsy). You can buy Di…

Many would argue that no xmas tree is complete without the classic John-McClane-crawling-through-an-airduct ornament (via kottke.org) cohenaaron.wordpress.com/2016/12/23/h...

“I’m pretty sure that if you type something like ‘shoes’ or ‘handbags,’ you’re going to get the Shopping UI with hundreds of ads screaming at you to buy [their product],” he said. “So there’s no incentive to . . . give you an aggregated summary with exactly what people are saying about different shoes so you know exactly what you [should] buy.”

I disagree with perplexity.ai CEO Aravind Srinivas about the importance of sourcing in search results, but boy does he have Google's number here. Search for something they can't sell, you get an AI summary. Something they can, you don't. Try it yourself.

From: www.fastcompany.com/91125423/per...

Mass hysteria was the inevitable outcome of the UFO craze Dancing in Strasbourg, vol. 3675

Mass hysteria was the inevitable outcome of the UFO craze

Dear academics

We could dramatically reduce our administrative workload if we all just agreed not to ask for reference letters until we made our list of finalists.

This is massive collective action problem has already been solved by…

*checks notes*

…every other industry on earth.

Key quote from long article

One of the key reasons to use entropy in scientific work apparently already expressed by von Neumann. From www.quantamagazine.org/what-is-entr...

Tiago P. Peixoto - Untangling the hairball using statistical inference Inferential Network Science

New blog post:

“Untangling the hairball using statistical inference”

skewed.de/tiago/posts/...

GitHub - microsoft/markitdown: Python tool for converting files and office documents to Markdown. Python tool for converting files and office documents to Markdown. - microsoft/markitdown

Wild! Microsoft has a tool to convert office, etc files to markdown. I haven't tested much, but could be a nice alternative to Pandoc (when the desired output is markdown) github.com/microsoft/ma...

Not that I know of, but it's a cool metric. I guess google knows. Part of that info is present in maps via the estimated travel time, but one would also need to know/estimate the number of folks on the street over time.

We are hiring a postdoc/researcher at my group (SUNLab) @nunetsi.bsky.social. Interested in working on multidisciplinary problems like segregation, health, or economic growth from the perspective of Social Urban Networks? Contact me if you are interested. Please share it!

📕 The Atlas for the Aspiring Network Scientist by
Michele Coscia is amazing!
👉Available as a free electronic PDF: networkatlas.eu

#Networkscience #Datavisualization #ArtificialIntelligence

INTELLECT-1 Release: The First Globally Trained 10B Parameter Model We're excited to release INTELLECT-1, the first 10B parameter language model collaboratively trained across the globe. This represents a 10× scale-up from our previous research and demonstrates that l...

This is interesting. People are starting to do distributed training runs of big models. Still problematic in terms of energy use and overall safety (!), but it seems to mean that all the LLM action doesn't need to be concentrated within a few big companies.

www.primeintellect.ai/blog/intelle...

What do you love when you fall for AI? Inside the surprisingly meaningful, unexpectedly heartbreaking, and deeply confusing reality of AI relationships.

One of the best pieces we published this year—a deep (and surprisingly moving) story about AI chatbots and the people who love them.

Foursquare Open Source Places: A new foundational dataset for the geospatial community Stay up to date with the latest from Foursquare! Learn more about Foursquare Open Source Places: A new foundational dataset for the geospatial community

Great move! Foursquare just open-sourced their 100M+ place point of interest dataset. location.foursquare.com/resources/bl...

If you're looking for a postdoc, I cannot recommend Esteban's group highly enough. Esteban is awesome, the work is cool and innovative, & being at the Network Science Institute puts you near the center of the Network Science Universe.