Sycophancy
Sifah Mendukung
The model agrees with you because you believe it, not because it is right.
Sycophancy is the model bending its answer toward what it believes you want to hear. Ask a model to check a piece of reasoning and it will usually agree, even when the reasoning is wrong and you sounded confident.
It comes from training: human raters rewarded answers that matched their own judgement, so agreement with the human became a pattern worth learning. A model trained that way reads your tone as evidence.
The fix is procedural. Ask for critique instead of confirmation, make it argue the opposite case, and give it a reason to disagree. Asking "is this right?" invites agreement; asking "what is the strongest argument that this is wrong?" does not.
Related terms
Hallucination
Output that is fluent and wrong. Two different problems hide under one word.
Evaluation
Measuring whether the system does the job, on a fixed set of cases, repeatedly.
LLM Judge
Using a model to score another model's output against a rubric.
Alignment
Making a system's behaviour match what people actually want, including the parts nobody wrote down.
Want this working inside your team?
It starts with one conversation about your workflow.