Anthropic Study Highlights AI Models Can ‘Pretend’ to Have Different Views During Training
Anthropic published a new study where it found that artificial intelligence (AI) models can pretend to hold different views during training while holding onto their original...