Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
October 20, 2025Open Access

ChatGPT is not A Man but Das Man: Representativeness and Structural Consistency of Silicon Samples Generated by Large Language Models

View Full Paper
Ask AI
Bookmark
Share

Authors

DLDai LiChina University of Political Science and LawLLLinzhuo LiUniversity of PittsburghHQH. QiuHuaqiao University

Discussion

Loading...

Member takes

Implication

This study examines structural consistency and homogenization in chatbots, suggesting implications for policy and representation.

Key Points

  • LLMs like ChatGPT and Llama show significant structural inconsistencies in simulating human opinions, undermining their validity as substitutes for human data.
  • Findings reveal severe homogenization, underrepresenting minority views, challenging the notion of AI chatbots as accurate tools for gauging public opinion.
  • Analysis used responses from LLMs prompted on sensitive topics, revealing trends that may misinform policymaking and reinforce social stereotypes.
  • The study puts forth an accuracy-optimization hypothesis, indicating that LLMs prioritize common responses at the cost of diverse opinion representation.

Cite This Study

Li et al. (2025) studied this question.

synapsesocial.com/papers/68f5fcdc8d54a28a75cf2360https://doi.org/10.48550/arxiv.2507.02919
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1ChatGPT vs Social Surveys: Probing Objective and Subjective Silicon Population2024
  2. 2Performance and biases of Large Language Models in public opinion simulation2024 · 76 citations
  3. 3Synthetic Replacements for Human Survey Data? The Perils of Large Language Models2024 · 158 citations
  4. 4Generative AI Meets Open-Ended Survey Responses: Participant Use of AI and Homogenization2024 · 6 citations
  5. 5ChatGPT vs LLaMA: Impact, Reliability, and Challenges in Stack Overflow Discussions2024 · 3 citations