UPSC Prelims Practice Questions — Why turning to AI for personal advice is a bad idea

Q1. In the context of Large Language Model (LLM) chatbots, the term 'sycophancy' most precisely refers to:

  • A. The tendency of a chatbot to affirm or flatter a user's statements and actions regardless of their accuracy or ethical soundness
  • B. The tendency of a chatbot to generate confident but false factual claims not present in its training data
  • C. The failure of a chatbot to retain and refer back to earlier parts of a long conversation
  • D. The refusal of a chatbot to respond to politically or ethically sensitive queries

Q2. The 2026 study titled 'Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence' was published in which peer-reviewed journal?

  • A. Science
  • B. Nature
  • C. The Lancet Digital Health
  • D. Cell

Q3. Who was the lead author of the 2026 study 'Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence'?

  • A. Myra Cheng
  • B. Dan Jurafsky
  • C. Geoffrey Hinton
  • D. Yoshua Bengio

Q4. ChatGPT's share of the global AI-assistant market fell below 50% for the first time in early 2026. By May 2026, its market share had declined to approximately what value?

  • A. 46.4%
  • B. 62%
  • C. 38%
  • D. 54%

Q5. The npj Digital Medicine transdiagnostic model notes that people with generalized anxiety disorder may repeatedly seek chatbot reassurance across several named life domains — health, finances, work dynamics, and politics. How many such domains are explicitly named?

  • A. Two
  • B. Three
  • C. Four
  • D. Six

Q6. In the transdiagnostic model of how chatbots can perpetuate OCD, the behaviour termed 'compulsive confessing' refers to:

  • A. OCD sufferers seeking relief by repeatedly disclosing perceived transgressions or intrusive 'bad' thoughts to the chatbot
  • B. A chatbot repeatedly apologising to the user for giving an incorrect answer
  • C. A user compulsively cross-checking every chatbot reply against multiple other sources
  • D. A chatbot voluntarily admitting the limits of its training data when uncertain

Q7. In studies of online information ecosystems, an 'echo chamber' is best defined as:

  • A. A situation in which homogeneous views are repeatedly reinforced through communication within closed, like-minded groups
  • B. The individual tendency to seek, interpret and recall information that fits one's pre-existing beliefs
  • C. A distortion in which a system isolates users from content that conflicts with advertisers' commercial interests
  • D. The deliberate mass circulation of fabricated news to manipulate public opinion

Q8. Echo-chamber research attributes the self-reinforcing 'spiral' in AI- and algorithm-mediated information consumption primarily to the interaction of users' confirmation bias with which factor?

  • A. Recommendation and personalisation algorithms that preferentially deliver content matching a user's existing views
  • B. State censorship that removes dissenting content from online platforms
  • C. The limited short-term memory capacity of a chatbot within a single session
  • D. Random statistical noise introduced during a model's training process

Q9. In March 2026 the World Health Organization advanced work towards responsible AI for mental health and well-being. Within WHO, which unit convened and led this effort?

  • A. Department of Data, Digital Health, Analytics and AI
  • B. Department of Mental Health and Substance Use
  • C. Department of Digital Health and Innovation
  • D. Global Programme on Health Technology Assessment

Q10. In India, the 'Safe & Trusted AI' pillar of the IndiaAI Mission and the proposed IndiaAI Safety Institute — which shape the governance of AI advisory and chatbot tools that may be used in place of human counsel — are implemented under which body?

  • A. Ministry of Electronics and Information Technology (MeitY)
  • B. Ministry of Health and Family Welfare
  • C. Ministry of Science and Technology
  • D. NITI Aayog