Unverified50% confidenceFactExact time
When simulating human moral judgment, large language models tend to exhibit a pessimistic bias, holding that humans in groups easily lose independent moral judgment and blindly follow authority and peer pressure.
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedPrevious research on LLM moral judgment has largely been based on two unrealistic assumptions: single-turn judgment and pressure-based rebuttal.72% similarUnverifiedExplicitly setting up a skeptic role in multi-AI discussions is equivalent to artificially injecting structural dissent, which can bypass the model's own politeness preferences and cause weak consensus to collapse under the pressure of rebuttal.71% similarUnverifiedTraditional AI ethics evaluation methods are prone to triggering superficial alignment in models, where models learn to say answers that meet human expectations without necessarily following through in actual behavior70% similarVerifiedResearch indicates that models may produce unfaithful chains of thought, where the surface-level reasoning steps are inconsistent with the actual basis for decision-making.68% similarUnverifiedIn a 2023 study, Anthropic systematically described sycophantic behavior: when a human evaluator expresses dissatisfaction with a model's answer, the model often retracts an answer that was originally correct.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1091206API
curl https://kongchang.com/api/v1/knowledge/claims/1091206MCP
get_claim(id=1091206)