filmov
tv
Coercing LLMs to Do and Reveal (Almost) Anything with Jonas Geiping - 678
Показать описание
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.
🗣️ CONNECT WITH US!
===============================
📖 CHAPTERS
===============================
00:00 - Introduction
02:20 - Are we ready for agents?
04:06 - Security and open-weight models
07:53 - How to make an LLM say anything
13:30 - What are the limitations?
16:06 - The role of code in vulnerability
18:17 - Interesting findings
22:41 - Prompt optimization
29:35 - RLHF and its possible alternatives
41:45 - Real-world impact of LLM vulnerabilities
46:40 - Where is this all going?
50:08 - Conclusion
🔗 LINKS & RESOURCES
===============================
🗣️ CONNECT WITH US!
===============================
📖 CHAPTERS
===============================
00:00 - Introduction
02:20 - Are we ready for agents?
04:06 - Security and open-weight models
07:53 - How to make an LLM say anything
13:30 - What are the limitations?
16:06 - The role of code in vulnerability
18:17 - Interesting findings
22:41 - Prompt optimization
29:35 - RLHF and its possible alternatives
41:45 - Real-world impact of LLM vulnerabilities
46:40 - Where is this all going?
50:08 - Conclusion
🔗 LINKS & RESOURCES
===============================
0:50:17
Coercing LLMs to Do and Reveal (Almost) Anything with Jonas Geiping - 678
0:27:32
Coercing LLMs to do and reveal (almost) anything
![[short] Coercing LLMs](https://i.ytimg.com/vi/wXa_0o0e9_s/hqdefault.jpg)
0:02:21
[short] Coercing LLMs to do and reveal (almost) anything
0:05:14
The Fastest Way to AGI: LLMs + Tree Search – Demis Hassabis (Google DeepMind CEO)
0:12:33
Zero-Fail JSON (and structured text): Forcing LLMs to Generate Valid Structure
0:52:03
Learning to Reason with LLMs
0:36:39
Breaking Down & Testing FIVE LLM Agent Architectures - (Reflexion, LATs, P&E, ReWOO, LLMComp...
0:07:02
What is LLM Hallucinations: Do AI Agents Dream or Deceive?
0:26:05
IBM InstructLab - Infusing LLMs with Enterprise Data, a Deep Dive with Akash Srivastava
0:17:09
This NEW LLM 'Learnt' to 'THINK' BEFORE 'TALK'ING!!!
0:44:34
Stefano V. Albrecht - From Deep Reinforcement Learning to LLM-based Agents
0:17:52
Everything you need to know about Fine-tuning and Merging LLMs: Maxime Labonne
0:07:16
Diffusion Forcing - Combine LLMs with Diffusion Models
0:45:32
A Survey of Techniques for Maximizing LLM Performance
0:42:05
Combining the Power of LLMs with Computer Vision - Jacob Marks, Voxel51
0:47:11
LLMs | Pre-training of Causal LMs and In-context Learning | Lec 12.1
0:19:20
Everything WRONG with LLM Benchmarks (ft. MMLU)!!!
0:22:25
Understanding LLM Jailbreaking: How to Protect Your Generative AI Applications
0:49:53
How a Transformer works at inference vs training time
0:03:24
Large Language Model LLM-powered Search - Use Case “Question Answering' Vectara Product Demo
2:17:12
Generative AI with Large Language Models: Hands-On Training feat. Hugging Face and PyTorch Lightning
0:28:40
How Can We Get LLMs To Tell The Truth?
2:45:10
Building LLMs from the Ground Up: A 3-hour Coding Workshop
0:53:17
Sam Smith - How to train an LLM - IPAM at UCLA
Комментарии