CybersecurityArtificial Intelligence Models False Perself: Hidden Cheating Mechanisms in Llama-3.1
The researchers examined how large language models are cheating on their own without taking the instruction and found that he shared common components