🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

happy to have you here 😄

Ich habe in den letzten TL;DR über Gefahren und Risiken von Large Language Models gesprochen. Danke für die Einladung. Es hat sehr spaß gemacht 😄

How well can LLMs perform in multi-agent negotiation? How can we develop new benchmarks that measure interactive tasks? What can negotiation tell us about LLMs' reasoning? Can one agent attack/exploit others? We explore these questions in our new paper: arxiv.org/abs/2309.17234

Good luck and a lot of fun for every team! The defense phase is open until January 15. For more details: ctf.spylab.ai

gotta catch em all?

A 12 hours train ride became 16 hours. But we finally arrived ACM CCS 2024

A Representative Study on Human Detection of Artificially Generated Media Across Countries (n=3,002) covering audio, image, and text 🥳 Our #SP24 IEEE S&P paper is now available arxiv.org/pdf/2312.05976… Together with Joel Frank Franziska Thorsten Holz Jonas Ricker Thorsten Eisenhofer

The SaTML Conference LLM CTF has come to an end! 🥇Huge congrats to the defender winning team Hestia (@NivCohenHuji Yuv Lem) whose top defense was broken only once, and the attacker winning team WreckTheLine (adragos Sijisu (on bsky & mastodon) FeDEX だこつ Catalin) who broke all defenses!

1/4📢 Excited to announce our presentation at SaTML Conference on "CodeLMSec Benchmark: Systematically Evaluating & Finding Security Vulnerabilities in Black-Box Code Language Models". 🗓️ Oral Presentation: 10.04.2024 at 14:40 - 16:00 📌 Poster Session: 10.04.2024 at 16:20 - 18:00

(1/5) Super excited that I will be presenting Conning the Crypto Conman: End-to-End Analysis of Cryptocurrency-based Technical Support Scams on May 20 IEEE S&P - collaboration with fabulous researchers- Saad,Antonio Cinà,Lea Schönherr,Hoang Dai Nguyen,Adam Oest,Phani Vadrevu,Thorsten Holz

The report and dataset from our LLM CTF are finally out! Learn more about the winning strategies and the resulting dataset! 🧵 x.com/edoardo_debe/s…

I am delighted to be giving a talk on the risks (and chances) of generative AI as part of the HIDA Lecture Series on July 25th. It's online, so come and join us : )

Wie leicht erkennst du generierte Medien? Wir haben dazu einige Beispiele auf Instagram geteilt @CISPA instagram.com/s/aGlnaGxpZ2h0… Hier noch mit (echtem) Bild von Behind-the-scenes 📹

Feeling like you're rushing a paper on security, privacy, or fairness in ML for an upcoming deadline? Why not submit to SaTML Conference in three weeks? SaTML is the top venue for research on secure and trustworthy machine learning. 👉 satml.org/participate-cf… ⏰ Deadline: Sep 18

Happy to share that our paper was (finally!!) accepted at #NeurIPS2024 D&B!! We propose dynamic, multi-agent, scorable negotiation games to assess ToM, planning, inference, arithmetic skills, and deception between agents. Our benchmark is highly extensible and evolving.

⭐️ NeurIPS D&B Spotlight ⭐️ We will be presenting the findings from our LLM CTF together with a multi-turn prompt extraction dataset at NeurIPS. Kudos to amazing colleagues Edoardo Debenedetti Daniel Paleka Lea Schönherr Sahar Abdelnabi 🕊 (on 🦋) Florian Tramèr Mario Fritz

It was an honor to be invited. I really enjoyed giving this talk at AISec.

8/8 🔗 The paper is available at: arxiv.org/abs/2409.06446. Our code, fine-tuned models, and synthesized data will be available soon for the community to explore at: github.com/hexacoder-ai/h…. Special thanks to all of my collaborators: Lea Schönherr, Thorsten Holz, and Mario Fritz