Picture for Ollie Jaffe

Ollie Jaffe

AI Sandbagging: Language Models can Strategically Underperform on Evaluations

Add code
Jun 12, 2024
Figure 1 for AI Sandbagging: Language Models can Strategically Underperform on Evaluations
Figure 2 for AI Sandbagging: Language Models can Strategically Underperform on Evaluations
Figure 3 for AI Sandbagging: Language Models can Strategically Underperform on Evaluations
Figure 4 for AI Sandbagging: Language Models can Strategically Underperform on Evaluations
Viaarxiv icon