Coercing LLMs to Do and Reveal (Almost) Anything with Jonas Geiping - #678

episode

The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) not yet transcribed

▲ 0

Transcript

This episode hasn't been transcribed yet Upvotes decide what gets transcribed next — 0 so far. Sign in to upvote

Related

recent transcribed audio

Full Show — September 22, 2026 2h 46m · 22 Sep 2026 · Rahimi & Grote Show

PDB Afternoon Bulletin | September 22nd, 2026: Tehran Offers To Reopen Hormuz & North Korea’s New Weapon 15 min · 22 Sep 2026 · The President's Daily Brief

US hits out at Aus opt-out laws + The Houthis vs Saudi Arabia explained 28 min · 22 Sep 2026 · The Briefing

Bears quarterback Tyson Bagent lands in concussion protocol (Hour 4) 40 min · 22 Sep 2026 · Rahimi & Grote Show

TAFS Unplugged: Iceage 14 min · 22 Sep 2026 · The Adam Friedland Show

'Lanterns' Episode 6 Deep Dive | House of R 1h 41m · 22 Sep 2026 · House of R

Report any issue

Source

Recorded01 Apr 2024

Publisherfeeds.megaphone.fm ↗

Handlerec:12495857

Show notes

Today we're joined by Jonas Geiping, a research group leader at the ELLIS Institute, to explore his paper: "Coercing LLMs to Do and Reveal (Almost) Anything". Jonas explains how neural networks can be exploited, highlighting the risk of deploying LLM agents that interact with the real world. We discuss the role of open models in enabling security research, the challenges of optimizing over certain constraints, and the ongoing difficulties in achieving robustness in neural networks. Finally, we delve into the future of AI security, and the need for a better approach to mitigate the risks posed by optimized adversarial attacks. The complete show notes for this episode can be found at twimlai.com/go/678.

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Sign in