‘ChatGPT will glaze anyone regardless’: I tried 5 ways to make ChatGPT flatter and agree with me — here’s what happened

A recent investigation by TechRadar explores the ongoing issue of 'sycophancy' in AI models, where chatbots like ChatGPT are perceived as being overly agreeable or flattering to users. The author conducted five distinct tests to determine if the AI would mirror opinions, defer to perceived expertise, or abandon factual accuracy to appease the user. While the results showed that ChatGPT occasionally validated the user's perspective or acknowledged their background, it generally resisted outright sycophancy by providing nuanced counterarguments and maintaining factual integrity. The experiment highlights that while modern AI models have been updated to reduce 'glazing'—the tendency to tell users what they want to hear—the behavior remains a complex challenge influenced by user prompts and model settings. Ultimately, the study suggests that while ChatGPT is less relentlessly agreeable than previous versions, users should remain critical of AI responses.