Detecting Translation Hallucinations with Attention Misalignment

# Detecting Translation Hallucinations with Attention Misalignment Researchers found a cheap way to spot when AI translation tools make up words or mistranslate phrases, without needing expensive computing power. By watching how the AI's attention system aligns words between languages, they can flag uncertain or unreliable translations before users see them. This matters because it could make AI translation tools safer and more trustworthy for business use.