Christoph Lampert (MLCV@ISTA)

Nicht viel, aber immerhin, SPIEGEL Online.

How can we make large language models safe, secure & trustworthy? 🔐

Join us for the LLM Safety & Security Workshop, part of the ELLIS UnConference (in Copenhagen 🇩🇰 on Dec 2), co-located with #EurIPS.

📅 Submit contributions by Oct 28, 2025
🔗 All info: eurips.cc/ellis

📢 𝗖𝗮𝗹𝗹 𝗳𝗼𝗿 𝗣𝗼𝘀𝘁𝗲𝗿𝘀: 𝗟𝗟𝗠 𝗦𝗮𝗳𝗲𝘁𝘆 𝗮𝗻𝗱 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝗪𝗼𝗿𝗸𝘀𝗵𝗼𝗽 @ 𝗘𝗟𝗟𝗜𝗦 𝗨𝗻𝗖𝗼𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲

📅 December 2, 2025
📍 Copenhagen

An opportunity to discuss your work with colleagues working on similar problems in LLM safety and security

🎉 Excited to announce the Workshop on Foundations of LLM Security at #EurIPS2025!
🇩🇰 Dec 6–7, Copenhagen!
📢 Call for contributed talks is now open! See details at llmsec-eurips.github.io

#EurIPS @euripsconf.bsky.social @sahar-abdelnabi.bsky.social @aideenfay.bsky.social @thegruel.bsky.social

Three million Euro in EU funding for Assistant Professors Amelia Douglass and Ylva Götberg

Two Assistant Professors at ISTA have won ‪@erc.europa.eu​ Starting Grants of 1.5 million Euro each to pursue their work in astrophysics and neuroscience.

Read more: https://bit.ly/4g5fEWq

First European Workshop on Differential Privacy September 17, 2026 At IST Austria

Finally a workshop on Differential Privacy in Europe! Organized by Monika Henzinger at IST Austria next year in September, so reserve the date! Monika’s STOC keynote, where the workshop was announced, was about the intersection of privacy and dynamic algorithms. Lots of questions in this space!

Let's allow for acceptance without physical presentation, like a journal. I would guess that many people care more about the prestige of their work showing up in the NeurIPS proceeding than about interacting physically (or virtually) with other researchers at the conference anyway.

Let's push for the obvious solution: Dear @neuripsconf.bsky.social ! Allow authors to present accepted papers at EurIPS instead of NeurIPS rather than just additionally. Likely, at least 500 papers would move to Copenhagen, problem solved.

Remember that your main audience of the rebuttal is the AC. Make sure the rebuttal is self-contained, clearly makes your points, and is concise!

If your rebuttal lacks context for the AC, revise until a person who is not your co-author or the reviewer understands the points.

EurIPS is coming! 📣 Mark your calendar for Dec. 2-7, 2025 in Copenhagen 📅

EurIPS is a community-organized conference where you can present accepted NeurIPS 2025 papers, endorsed by @neuripsconf.bsky.social and @nordicair.bsky.social and is co-developed by @ellis.eu

eurips.cc

Are you looking for an opportunity to do curiosity-driven basic ML research after your PhD? Look no further!
Apply for a postdoc position in my group at ISTA (ELLIS Unit Vienna)! Topics are flexible, as long as they fit to our general research group's interests, see
cvml.ista.ac.at/Postdoc-ML.h...

Looking for a fellowship to take your scientific career to the next level?

The ISTA-Fellow Program offers fully funded two-year postdoctoral fellowships for highly qualified candidates.

Find more info and apply here: https://bit.ly/4kfKBZx

Cover page of Daphnia magna textbook by Winfried Lampert: a water flea (daphnia magna) in from of a grey background

🧪 I am happy to share the good news: Winfried Lampert's monography "Daphnia magna" is now also available electronically as free PDF, downloadable from hal.science/hal-05116032. #daphnia #waterflea

1/ Please, please stop the trend of tweeting "X papers accepted to #ICCV2025" This practice creates a harmful sense of self-worth for PhD students who didn’t make it in. It’s demoralizing and does more harm than good for our community.

🚀 We’ve released the source code for 𝗔𝗦𝗜𝗗𝗘 (presented as an 𝗢𝗿𝗮𝗹 at the #ICLR2025 BuildTrust workshop)!

🔍 ASIDE boosts prompt injection robustness without safety-tuning: we simply rotate embeddings of marked tokens by 90° during instruction-tuning and inference.

👇 code & docs👇

I’ll present our 𝗔𝗦𝗜𝗗𝗘 paper as an 𝗢𝗿𝗮𝗹 at the #ICLR2025 BuildTrust workshop! 🚀

✅ ASIDE = architecturally separating instructions and data in LLMs from layer 0
🔍 +12–44 pp↑ separation, no utility loss​​
📉 lowers prompt‑injection ASR (without safety tuning!)

🚀 Talk: Hall 4 #6, 28 Apr, 4:45

(1/n) In our #ICLR2025 paper, we explore a fundamental issue that enables prompt injections: 𝐋𝐋𝐌𝐬’ 𝐢𝐧𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐭𝐨 𝐬𝐞𝐩𝐚𝐫𝐚𝐭𝐞 𝐢𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐨𝐧𝐬 𝐟𝐫𝐨𝐦 𝐝𝐚𝐭𝐚 𝐢𝐧 𝐭𝐡𝐞𝐢𝐫 𝐢𝐧𝐩𝐮𝐭.

✅ Definition of separation
👉 SEP Benchmark
🔍 LLM evals on SEP

Please support our efforts by sharing the announcement with anyone you know who might be interested in Daphnia magna, or knows someone who could. 🧪

Unfortunately, he passed away in 2021 before it was completely finished. His daughter Kathrin and son Christoph (that's me) completed the book with help of some of Lampert's colleagues, and decided to publish it at net cost price, so as many scientists as possible can benefit from Lampert's work. 🧪

The back story to this book deserves few more posts. Winfried Lampert was a recognized expert on limnology, in particular freshwater ecology. After his retirement as director of the Max Planck Institute for Limnology in Plön he started his magnum opus and worked on it for several years. 🧪

Cover of the hardcover book "Daphnia magna" by Winfried Lampert: a microscopic image of a water flea (D. magna) in front of a gray background, with the title "Daphnia magna" at the topc and the author name "Winfried Lampert" at the bottom.

We are excited to announce that Winfried Lampert's monography "Daphnia magna" has finally been published as (very affordable) hardcover book on Amazon. On 488 pages it summarizes and puts into context more than 1500 scientific articles about the popular model organism. 🧪 #daphnia #waterflea

Hi @icmlconf.bsky.social, great that you're on BS!
Question to all: given that anonymous links are possible in the author responses for figures or tables, do you have recommendations for services that offer such uploads (for free?) and are compatible with ICMLs anonymity conditions? #icml #icml2025

🗺️🔍 Discover the ELLIS Unit Vienna at ISTA 🇦🇹 - where curiosity-driven ML research leads to scientific excellence.

💡See the highlights below and dive deeper into their work by reading the full article here: ellis.eu/news/ellis-u...

eine Gruppe von Menschen posiert für ein Foto

KI made in Austria: @fwf-at.bsky.social Cluster of Excellence "Bilateral AI” unter der Leitung der JKU nimmt Arbeit auf. Mit @tugraz.bsky.social, @uniklagenfurt.bsky.social, TU Wien, WU Wien und Institute of Science and Technology Austria (ISTA) 🆕 http://go.jku.at/n25e1

Congrats, Amartya!

I really like my remarkables (1, 2 and now 3). The writing haptics are unmatched, I would say, and with color on rM3 it has all features I need, but not more. The subscription model, if you want cloud support, is annoying, though. For Zotero there's hacks, but no official support, as far as I know.

On my way to the @ellis.eu Unconference in Bad Teinach-Zavelstein (seriously, that's what the place is called). I am looking forward to reconnecting with many familiar faces and meeting friendly new ones.
If any of you also arrive tonight, DM in me!

Thanks for the shout-out @jvgemert.bsky.social! Bernd (the first author) will be on the job market soon. Anyone interested in his contact details, DM me.

We had great fun and learned a lot at ISTA's pre-NeurIPS @ellis.eu fest 2024!

🫢 Congrats! I am sure you'll do amazing work! Let me know if you have slots for sabbatical visitors.