Picture for Charlie Rogers-Smith

Charlie Rogers-Smith

Unelicitable Backdoors in Language Models via Cryptographic Transformer Circuits

Add code
Jun 03, 2024
Viaarxiv icon

BadLlama: cheaply removing safety fine-tuning from Llama 2-Chat 13B

Add code
Oct 31, 2023
Figure 1 for BadLlama: cheaply removing safety fine-tuning from Llama 2-Chat 13B
Figure 2 for BadLlama: cheaply removing safety fine-tuning from Llama 2-Chat 13B
Figure 3 for BadLlama: cheaply removing safety fine-tuning from Llama 2-Chat 13B
Figure 4 for BadLlama: cheaply removing safety fine-tuning from Llama 2-Chat 13B
Viaarxiv icon

LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B

Oct 31, 2023
Figure 1 for LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B
Figure 2 for LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B
Figure 3 for LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B
Figure 4 for LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B
Viaarxiv icon

Approximate Bayesian Computation via Population Monte Carlo and Classification

Oct 29, 2018
Figure 1 for Approximate Bayesian Computation via Population Monte Carlo and Classification
Figure 2 for Approximate Bayesian Computation via Population Monte Carlo and Classification
Figure 3 for Approximate Bayesian Computation via Population Monte Carlo and Classification
Figure 4 for Approximate Bayesian Computation via Population Monte Carlo and Classification
Viaarxiv icon