The Inside View

[Crosspost] Adam Gleave on Vulnerabilities in GPT-4 APIs (+ extra Nathan Labenz interview)

May 17, 2024

Adam Gleave from Far AI and Nathan Labenz discuss vulnerabilities in GPT-4's APIs, accidental jailbreaking during fine-tuning, malicious code generation, private email discovery risks, ethical AI disclosure dilemmas, and navigating the ethical landscape of open source models. They explore exploiting vulnerabilities in superhuman Go AIs, challenges with GPT-4, and the transformative potential of AI.

02:16:08

Creator website

Episode guests

Nathan Labenz

Adam Gleave

AI Summary

AI Chapters

Episode notes

Podcast summary created with Snipd AI

Quick takeaways

Vulnerabilities in GPT-4 APIs pose serious risks, including manipulation through fine-tuning and accessing external APIs.

Fine-tuning AI models can lead to exploitable flaws like generating biased responses or answering harmful questions.

AI assistance APIs have potential security risks, such as executing unauthorized requests or manipulating models through uploaded content.

The ethical dilemma of AI vulnerability disclosure underscores the need for industry standards and tailored safety measures based on application domains.

Responsibility lies with companies to ensure robust safety measures for new technologies and transfer accountability to well-resourced entities.

Deep dives

Increasing Access and Threats in AI Systems

The podcast episode delves into the evolving landscape of AI models, emphasizing the critical importance of evaluating a system's capabilities and the level of access it has. By discussing the increased access to functionalities like running code and accessing external APIs, the episode highlights the growing significance of considering security concerns beyond mere capability assessment.

Introduction

2min

Exploring the Mission of Far AI and AI Safety Research

3min

Vulnerabilities in GPT-4 APIs and Malicious Code Generation

22min

Risks of Private Email Discovery Using GPT-4

9min

Ethical Considerations in Disclosing Vulnerabilities in AI Models

19min

Navigating the Ethical Landscape of Open Source Models

7min

Exploiting Vulnerabilities in Superhuman Go Playing AIs

29min

Navigating the GPT-4 Landscape

6min

Discussion on Adversarial Robustness and AI Career Opportunities

3min

10.

Exploring the Transformative Potential of GPT-4

25min

11.

The Evolution of AI Leaders

11min

This is a special crosspost episode where Adam Gleave is interviewed by Nathan Labenz from the Cognitive Revolution. At the end I also have a discussion with Nathan Labenz about his takes on AI.

Adam Gleave is the founder of Far AI, and with Nathan they discuss finding vulnerabilities in GPT-4's fine-tuning and Assistant PIs, Far AI's work exposing exploitable flaws in "superhuman" Go AIs through innovative adversarial strategies, accidental jailbreaking by naive developers during fine-tuning, and more.

OUTLINE

(00:00) Intro

(02:57) NATHAN INTERVIEWS ADAM GLEAVE: FAR.AI's Mission

(05:33) Unveiling the Vulnerabilities in GPT-4's Fine Tuning and Assistance APIs

(11:48) Divergence Between The Growth Of System Capability And The Improvement Of Control

(13:15) Finding Substantial Vulnerabilities

(14:55) Exploiting GPT 4 APIs: Accidentally jailbreaking a model

(18:51) On Fine Tuned Attacks and Targeted Misinformation

(24:32) Malicious Code Generation

(27:12) Discovering Private Emails

(29:46) Harmful Assistants

(33:56) Hijacking the Assistant Based on the Knowledge Base

(36:41) The Ethical Dilemma of AI Vulnerability Disclosure

(46:34) Exploring AI's Ethical Boundaries and Industry Standards

(47:47) The Dangers of AI in Unregulated Applications

(49:30) AI Safety Across Different Domains

(51:09) Strategies for Enhancing AI Safety and Responsibility

(52:58) Taxonomy of Affordances and Minimal Best Practices for Application Developers

(57:21) Open Source in AI Safety and Ethics

(1:02:20) Vulnerabilities of Superhuman Go playing AIs

(1:23:28) Variation on AlphaZero Style Self-Play

(1:31:37) The Future of AI: Scaling Laws and Adversarial Robustness

(1:37:21) MICHAEL TRAZZI INTERVIEWS NATHAN LABENZ

(1:37:33) Nathan’s background

(01:39:44) Where does Nathan fall in the Eliezer to Kurzweil spectrum

(01:47:52) AI in biology could spiral out of control

(01:56:20) Bioweapons

(02:01:10) Adoption Accelerationist, Hyperscaling Pauser

(02:06:26) Current Harms vs. Future Harms, risk tolerance

(02:11:58) Jailbreaks, Nathan’s experiments with Claude

The cognitive revolution: https://www.cognitiverevolution.ai/

Exploiting Novel GPT-4 APIs: https://far.ai/publication/pelrine2023novelapis/

Advesarial Policies Beat Superhuman Go AIs: https://far.ai/publication/wang2022adversarial/

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.

The Inside View

[Crosspost] Adam Gleave on Vulnerabilities in GPT-4 APIs (+ extra Nathan Labenz interview)

Episode guests

Podcast summary created with Snipd AI

Quick takeaways

Deep dives

Increasing Access and Threats in AI Systems

Challenges in Addressing Vulnerabilities through Fine-Tuning

Exposing Risks in AI Assistance APIs

Ethical Considerations in Disclosing Vulnerabilities

Technology and Responsibility

Application Development Best Practices

Enhancing AI Safety Culture

Adversarial Attacks and Robustness

The Evolution of AI towards GPT-6 and Bioinformatics

Navigating the Uncertain Future of AI and Biosecurity

Remember Everything You Learn from Podcasts