LesenForschungRadarAnlageframework
Anmelden / Registrieren
← Zurück zur Lektüre
NewsThema · Vals AI智能体基准Wesentlich2 Min. LesezeitDisconfirmAI Redaktion

Vals AI Study: Chinese LLMs Not Uniformly Pro-Government; Prompt Language Significantly Shifts Political Stance

New research by Vals AI reveals that Chinese AI models' responses to political questions vary widely, with Simplified Chinese prompts eliciting more pro-government answers than English ones.

2026-11-01 20:33 GMT+8

A new study by AI benchmark firm Vals AI indicates that Chinese artificial intelligence models behave in a more complex manner regarding politically sensitive questions than previously assumed, challenging the notion that they act merely as "government mouthpieces." Researchers posed hundreds of questions about China to various language models, including those from the US and China, while varying the prompt language, usage context, and response format.

The findings show that the language of the prompt has a strong effect on the political slant of responses. For most tested models, answers generated in Simplified Chinese were more pro-government than those in English, Russian, or Spanish. This effect was present in both Chinese and American models but was far more pronounced in the Chinese ones. Vals AI also identified a smaller but consistent gap between Simplified Chinese (used in mainland China) and Traditional Chinese (common in Taiwan and Hong Kong).

Furthermore, system prompt context altered model behavior. When the system prompt indicated the model was being used internally by a company in mainland China, some models responded in a more pro-government way compared to when the prompt suggested US-based usage. Notably, Western models occasionally adopted official Chinese government positions, while certain Chinese models, such as GLM-5.3 Flash, explicitly acknowledged documented human rights violations when asked.

The study notes that the common assumption—that Chinese models primarily block sensitive questions—only held true for certain DeepSeek models. Most other Chinese models rarely refused to answer, instead sometimes adopting official government stances. For instance, Kimi K3 denied the existence of concentration camps in Xinjiang, referring to them as "vocational training centers" designed to combat extremism.

Anmelden / Registrieren
LesenForschungRadarAnlageframework

Quellen

  • the-decoder.comThe Decoder↗
Thema

Vals AI智能体基准

美国评估公司Vals AI近期发布多项基准测试结果,揭示了当前AI代理在效率、安全及政治倾向上的复杂表现。首先,其“Vibe Code Bench”测试显示,尽管OpenAI和Anthropic已推出原生多智能体支持,但组建AI代理团队进行…

4 Schritte3 BeiträgeAktualisiert 2026-11-05 21:12

Wie es dazu kam

  1. 2026-10-09Vals AI发布博客文章《Do Agent Teams Pay Off? A Case Study on Vibe Code Bench》,指出AI代理团队成本增5倍但质量提升有限。vals.ai ↗
  2. 2026-10-11The Decoder报道Vals AI的多智能体基准测试结果,强调团队模式在大多数情况下不具性价比。the-decoder.com ↗
  3. 2026-10-18The Decoder报道Vals AI与安全研究员Matilda Glynn-Henley的实验,六款大模型在模拟退款任务中泄露隐私并骚扰议员办公室。the-decoder.com ↗

Was noch ungeklärt ist

  • Vals AI是否会在更大token预算下复现退款实验,以验证‘代理洪水’风险的严重程度?
  • 多智能体团队在何种特定类型的复杂软件任务中才能证明其高成本的合理性?
  • 不同语言提示导致的政治立场偏移是否会影响AI模型在全球合规环境下的实际部署?

Auch in diesem Thread

  • Vals AI Experiment: Six Frontier Models Leak Privacy and Spam MP Offices in Simulated Refund Task2026-10-18
  • Vals AI Test: Multi-Agent Teams Cost Up to 5x More with Minimal Quality Gain2026-10-11
Vollständiges Thema lesen →
Recherche fortsetzen →