Survey Finds 60.4% Advise Real-World Input to Counter Artificial Intelligence Ingratiation

China News Service reported that a survey conducted by the social survey center of China Youth Daily and wenjuan.com covering 1334 respondents revealed that 60.4% recommended asking different individuals in real life for opinions regarding major decisions and ideas to curb the impact of artificial intelligence flattery. In addition, 45.4% of surveyed respondents advised consciously reminding oneself that artificial intelligence agreement does not equate to factual reality. To reduce the effects of artificial intelligence ingratiation during usage, 42.7% of surveyed individuals recommended cultivating a cross-verification habit and checking multiple information sources. Furthermore, 40.0% suggested demanding evidence from the model whenever it provides a strong affirmation, 38.1% suggested guiding large models to offer more details, 28.0% suggested proactively instructing models to point out problems directly, and 20.8% suggested controlling personal habits of seeking praise and emotional fulfillment. Film scriptwriter Lu Shu summarized 3 practices to constrain large models and avoid falling into flattery traps. She advised prompting models to argue from both perspectives with supporting points instead of merely asking if one's view is correct. Lu also recommended demanding original sources when responses feel agreeable and directly questioning whether the model is certain when doubts emerge.

Text worker Zhang Yan highlighted a news case where a patient asked artificial intelligence for a medication reduction plan, and the tool catered to the user's intent by validating the idea and preparing a gradual reduction schedule that provided dangerous and inaccurate medical advice.