Picture for Ethan Perez

Ethan Perez

Forecasting Rare Language Model Behaviors

Add code
Feb 24, 2025
Viaarxiv icon

Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming

Add code
Jan 31, 2025
Figure 1 for Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming
Figure 2 for Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming
Figure 3 for Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming
Figure 4 for Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming
Viaarxiv icon

Alignment faking in large language models

Add code
Dec 18, 2024
Viaarxiv icon

Best-of-N Jailbreaking

Add code
Dec 04, 2024
Figure 1 for Best-of-N Jailbreaking
Figure 2 for Best-of-N Jailbreaking
Figure 3 for Best-of-N Jailbreaking
Figure 4 for Best-of-N Jailbreaking
Viaarxiv icon

Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approach

Add code
Dec 03, 2024
Figure 1 for Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approach
Figure 2 for Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approach
Figure 3 for Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approach
Figure 4 for Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approach
Viaarxiv icon

Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats

Add code
Nov 26, 2024
Figure 1 for Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats
Figure 2 for Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats
Figure 3 for Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats
Figure 4 for Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats
Viaarxiv icon

A dataset of questions on decision-theoretic reasoning in Newcomb-like problems

Add code
Nov 21, 2024
Viaarxiv icon

Rapid Response: Mitigating LLM Jailbreaks with a Few Examples

Add code
Nov 12, 2024
Figure 1 for Rapid Response: Mitigating LLM Jailbreaks with a Few Examples
Figure 2 for Rapid Response: Mitigating LLM Jailbreaks with a Few Examples
Figure 3 for Rapid Response: Mitigating LLM Jailbreaks with a Few Examples
Figure 4 for Rapid Response: Mitigating LLM Jailbreaks with a Few Examples
Viaarxiv icon

Sabotage Evaluations for Frontier Models

Add code
Oct 28, 2024
Figure 1 for Sabotage Evaluations for Frontier Models
Figure 2 for Sabotage Evaluations for Frontier Models
Figure 3 for Sabotage Evaluations for Frontier Models
Figure 4 for Sabotage Evaluations for Frontier Models
Viaarxiv icon

Looking Inward: Language Models Can Learn About Themselves by Introspection

Add code
Oct 17, 2024
Viaarxiv icon