Picture for Aleksandar Petrov

Aleksandar Petrov

Mimicking User Data: On Mitigating Fine-Tuning Risks in Closed Large Language Models

Add code
Jun 12, 2024
Viaarxiv icon

Universal In-Context Approximation By Prompting Fully Recurrent Models

Add code
Jun 03, 2024
Viaarxiv icon

Near to Mid-term Risks and Opportunities of Open Source Generative AI

Add code
Apr 25, 2024
Viaarxiv icon

Prompting a Pretrained Transformer Can Be a Universal Approximator

Add code
Feb 22, 2024
Figure 1 for Prompting a Pretrained Transformer Can Be a Universal Approximator
Figure 2 for Prompting a Pretrained Transformer Can Be a Universal Approximator
Figure 3 for Prompting a Pretrained Transformer Can Be a Universal Approximator
Figure 4 for Prompting a Pretrained Transformer Can Be a Universal Approximator
Viaarxiv icon

When Do Prompting and Prefix-Tuning Work? A Theory of Capabilities and Limitations

Add code
Oct 30, 2023
Viaarxiv icon

The ARRT of Language-Models-as-a-Service: Overview of a New Paradigm and its Challenges

Add code
Sep 28, 2023
Viaarxiv icon

Language Model Tokenizers Introduce Unfairness Between Languages

Add code
May 17, 2023
Figure 1 for Language Model Tokenizers Introduce Unfairness Between Languages
Figure 2 for Language Model Tokenizers Introduce Unfairness Between Languages
Figure 3 for Language Model Tokenizers Introduce Unfairness Between Languages
Figure 4 for Language Model Tokenizers Introduce Unfairness Between Languages
Viaarxiv icon

Certifying Ensembles: A General Certification Theory with S-Lipschitzness

Add code
Apr 25, 2023
Figure 1 for Certifying Ensembles: A General Certification Theory with S-Lipschitzness
Figure 2 for Certifying Ensembles: A General Certification Theory with S-Lipschitzness
Figure 3 for Certifying Ensembles: A General Certification Theory with S-Lipschitzness
Figure 4 for Certifying Ensembles: A General Certification Theory with S-Lipschitzness
Viaarxiv icon

HiddenGems: Efficient safety boundary detection with active learning

Add code
Oct 25, 2022
Figure 1 for HiddenGems: Efficient safety boundary detection with active learning
Figure 2 for HiddenGems: Efficient safety boundary detection with active learning
Figure 3 for HiddenGems: Efficient safety boundary detection with active learning
Figure 4 for HiddenGems: Efficient safety boundary detection with active learning
Viaarxiv icon

Robustness of Unsupervised Representation Learning without Labels

Add code
Oct 08, 2022
Figure 1 for Robustness of Unsupervised Representation Learning without Labels
Figure 2 for Robustness of Unsupervised Representation Learning without Labels
Figure 3 for Robustness of Unsupervised Representation Learning without Labels
Figure 4 for Robustness of Unsupervised Representation Learning without Labels
Viaarxiv icon