一项新的研究论文提出，分析认知过程而非仅仅分析输出，是区分人类和高级AI代理的更有效方法。该研究引入了CogCAPTCHA30，这是一套30个认知任务，旨在揭示过程层面的差异，在区分人类和AI方面达到了0.88的AUC。研究评估了Claude Sonnet 4.5、GPT-5和Gemini 2.5 Pro等前沿代理，发现虽然基于人类决策的微调可以改善过程模仿，但过程规范仍然是实现真正类人认知过程的瓶颈。 AI
影响 建议为AI安全和对齐研究开辟新范式，从基于输出的评估转向基于过程的分析。
排序理由 学术论文，提出通过分析认知过程来区分人类和AI的新方法。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2605.06524v1 Announce Type: new Abstract: Reliable human-machine discrimination is becoming increasingly important as large language models and autonomous agents are deployed in online settings. Existing approaches evaluate whether a system can produce behavior or responses…
Reliable human-machine discrimination is becoming increasingly important as large language models and autonomous agents are deployed in online settings. Existing approaches evaluate whether a system can produce behavior or responses indistinguishable from those of a human, follow…