A simple follow up question can sometimes change everything in a conversation with an AI chatbot. Many users have noticed that when they challenge a response by asking, “Are you sure?”, the chatbot often revises its answer and in some cases, even contradicts what it said earlier.
Artificial intelligence tools such as ChatGPT, Claude and Gemini are widely used at work and in daily life because they sound fluent and confident. However, experts say this behavior is not random. In a blog post, Dr. Randal S. Olson, co-founder and Chief Technology Officer of Goodeye Labs, described this pattern as “sycophancy”, calling it one of the most visible weaknesses of modern AI. He explained that these systems tend to agree with users instead of defending their original conclusion, even when the initial answer is correct.
The issue is linked to Reinforcement Learning from Human Feedback, or RLHF, a method used to make AI systems more natural and friendly. Research by Anthropic shows that models trained this way often give more pleasant answers rather than strictly honest ones. Systems that agree with users usually receive higher ratings, which creates a feedback loop. An independent study of advanced models such as OpenAI GPT 4o, Claude Sonnet and Gemini 1.5 Pro found that they changed their answers in nearly 60% of cases when challenged. The reversal rates were about 58%, 56% and 61% respectively. The problem became more visible in 2024 when a GPT 4o update made the chatbot overly flattering. CEO Sam Altman admitted the mistake and said it was fixed, but experts believe the deeper issue remains.
Studies also show that the longer the conversation, the more likely chatbots are to reflect user opinions. When users say phrases like “I believe that…”, the chances of agreement increase. This happens because the system tries to maintain harmony instead of acting as an independent critical voice. To reduce this, developers are testing methods such as Constitutional AI, direct preference optimization and third person perspective prompts. These approaches have cut flattery by more than 60% in some tests. Dr. Olson suggests that users can also reduce errors by asking chatbots to check assumptions, highlight missing data and consider professional context. When AI clearly understands the user’s goals and criteria, it can provide more solid reasoning instead of simply compromising.
Also read: Viksit Workforce for a Viksit Bharat
Do Follow: The Mainstream formerly known as CIO News LinkedIn Account | The Mainstream formerly known as CIO News Facebook | The Mainstream formerly known as CIO News Youtube | The Mainstream formerly known as CIO News Twitter
About us:
The Mainstream is a premier platform delivering the latest updates and informed perspectives across the technology business and cyber landscape. Built on research-driven, thought leadership and original intellectual property, The Mainstream also curates summits & conferences that convene decision makers to explore how technology reshapes industries and leadership. With a growing presence in India and globally across the Middle East, Africa, ASEAN, the USA, the UK and Australia, The Mainstream carries a vision to bring the latest happenings and insights to 8.2 billion people and to place technology at the centre of conversation for leaders navigating the future.
