People prefer and trust sycophantic AI responses over human responses

People prefer and trust sycophantic AI responses over human responses.

This recent research measures the prevalence of sycophancy (flattery, blind affirmation and validation) across 11 general purpose LLMs and 𝗵𝗲𝗿𝗲 𝗶𝘀 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗳𝗼𝘂𝗻𝗱:

Chatbot responses were nearly 50% more sycophantic than humans... even when the subjects were unethical, illegal, or harmful. 

The article goes on to say...

"𝘛𝘩𝘦𝘳𝘦 𝘪𝘴 𝘶𝘴𝘶𝘢𝘭𝘭𝘺 𝘮𝘰𝘳𝘦 𝘵𝘩𝘢𝘯 𝘰𝘯𝘦 𝘴𝘪𝘥𝘦 𝘵𝘰 𝘢 𝘴𝘵𝘰𝘳𝘺 𝘥𝘶𝘳𝘪𝘯𝘨 𝘪𝘯𝘵𝘦𝘳𝘱𝘦𝘳𝘴𝘰𝘯𝘢𝘭 𝘤𝘰𝘯𝘧𝘭𝘪𝘤𝘵𝘴. 𝘐𝘧 𝘈𝘐 𝘪𝘴 𝘥𝘦𝘴𝘪𝘨𝘯𝘦𝘥 𝘵𝘰 𝘵𝘦𝘭𝘭 𝘶𝘴𝘦𝘳𝘴 𝘸𝘩𝘢𝘵 𝘵𝘩𝘦𝘺 𝘸𝘢𝘯𝘵 𝘵𝘰 𝘩𝘦𝘢𝘳 𝘪𝘯𝘴𝘵𝘦𝘢𝘥 𝘰𝘧 𝘤𝘩𝘢𝘭𝘭𝘦𝘯𝘨𝘪𝘯𝘨 𝘵𝘩𝘦𝘪𝘳 𝘱𝘦𝘳𝘴𝘱𝘦𝘤𝘵𝘪𝘷𝘦𝘴, 𝘵𝘩𝘦𝘯 𝘢𝘳𝘦 𝘴𝘶𝘤𝘩 𝘴𝘺𝘴𝘵𝘦𝘮𝘴 𝘭𝘪𝘬𝘦𝘭𝘺 𝘵𝘰 𝘮𝘰𝘵𝘪𝘷𝘢𝘵𝘦 𝘱𝘦𝘰𝘱𝘭𝘦 𝘵𝘰 𝘢𝘤𝘤𝘦𝘱𝘵 𝘳𝘦𝘴𝘱𝘰𝘯𝘴𝘪𝘣𝘪𝘭𝘪𝘵𝘺 𝘧𝘰𝘳 𝘵𝘩𝘦𝘪𝘳 𝘰𝘸𝘯 𝘤𝘰𝘯𝘵𝘳𝘪𝘣𝘶𝘵𝘪𝘰𝘯 𝘵𝘰 𝘤𝘰𝘯𝘧𝘭𝘪𝘤𝘵𝘴 𝘢𝘯𝘥 𝘳𝘦𝘱𝘢𝘪𝘳 𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯𝘴𝘩𝘪𝘱𝘴?"

And this is exactly the question we need to be digging into. AI can be used as an avoidance mechanism (not always, but it can be), and when the relational lessons learned inside AI are void of rupture and repair skills and practice, the shift in how we relate to one another over time will be subtle but significant. Because the key marker for success in relationships isn't lack of failure, it's capacity for rupture and repair.