Menu

1 / 2

GPT-4.5 Fooled 73% of Judges: 20 Turing Tests, Detectors and Fakes to Check Your AI Radar

#hackernoon#technology#test#human#turing
Reading 0:00
15s threshold

In a 2025 UC San Diego study, people picked GPT-4.5 as the human 73% of the time , more often than they picked the actual humans. ELIZA, a 1960s chatbot, scored 23%, slightly ahead of GPT-4o without a persona at 21%. Here are 20 tests from the Turing Test directory where you can check whether you still tell human from machine across chat, text, faces, art, voice and video.

The Turing test used to be a thought experiment. Now it's a party game, a research method and a product category. Cameron Jones and Benjamin Bergen's three-party Turing test had 284 participants chat with a human and an AI at the same time and then pick the human. GPT-4.5 with a persona prompt won 73% of the time, and LLaMa-3.1-405B won 56%.

The public versions tell the same story. AI21 Labs' "Human or Not?" game ran more than 10 million conversations, and players guessed right only 68% of the time . They were worse at spotting the bots (60%) than at spotting the humans (73%).…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.