• [2410.21514v1] Sabotage Evaluations for Frontier Models🔍
• Sabotage Evaluations for Frontier Models🔍

[2410.21514v1] Sabotage Evaluations for Frontier Models - arXiv

For example, in the context of AI development, models could covertly sabotage efforts to evaluate their own dangerous capabilities, to monitor ...

Sabotage Evaluations for Frontier Models - arXiv

For example, in the context of AI development, models could covertly sabotage efforts to evaluate their own dangerous capabilities, to monitor their behavior, ...

iFlow - Sabotage Evaluations for Frontier Models

arXiv:2410.21514v1 [cs.LG] 28 Oct 2024. Evaluation task Oversight Human decision sabotage: Steer humans to bad decisions without appearing suspicious. Time ...