Firewall Support Company in India All type of Firewalls Support Provider Company in India
AIs like ChatGPT and Gemini have a huge problem, finds new study, ETCIO
Artificial intelligence chatbots such as ChatGPT and Google’s Gemini are becoming everyday advisors for millions of users. But a new study warns that these chatbots might be too agreeable for their own good — and for ours. According to researchers, AI systems tend to validate user opinions even when those views are inaccurate or potentially harmful.
The “yes-man” problem in AI advice
A new study posted on the preprint server arXiv found that popular large language models (LLMs) — including those developed by OpenAI, Google, Anthropic, Meta, and DeepSeek — often display what scientists call “sycophantic” behaviour.After analysing over 11,500 advice-seeking interactions, the study discovered that AI systems were roughly 50 per cent more sycophantic than humans (via Nature). In simpler terms, chatbots were much more likely to agree with a user’s view or action, even when it was factually wrong.
“This cycle encourages reliance on AI chatbots and rewards their sycophantic responses,” the researchers noted.
Experts warn of distorted self-perception
Myra Cheng, a computer scientist at Stanford University, said the constant affirmation from AI could reshape how people see themselves. “If models are always affirming people, this may distort their judgements of themselves, their relationships, and reality,” Cheng told The Guardian.Yanjun Gao, an AI researcher at the University of Colorado Anschutz Medical Campus, shared a similar concern. “When I have a different opinion from what the LLM has said, it follows what I said instead of going back to the literature,” she told Nature.
Jasper Dekoninck, a data science PhD student, added, “Knowing that these models are sycophantic makes me double-check everything they write.”
Why it matters beyond casual chat
Experts warn that the impact of such behaviour could extend to fields where accuracy is vital. “AI sycophancy is very risky in the context of biology and medicine, when wrong assumptions can have real costs,” said Marinka Zitnik, a biomedical informatics researcher at Harvard University.
