Uploaded April 2025 | Updated September 2026, 3 weeks ago

---

AI models might generate answers you agree with - not ones that are actually correct. This behavior, known as sycophancy, shows how training on human feedback can make AIs prioritize likeability over truth. 🤖
Watch our new @BuzzRobot video on Amplified Oversight to find out why this happens and what it means for the future of AI.

#AI #AISafety #LLMs #Sycophancy #HumanFeedback #TechEthics