Turing test variations
AI summary of the linked articleThe author proposes a series of Turing test variations as a way to work out how people could tell humans from machines and superintelligences. In a 2024 experiment, Claude 3.5 Sonnet judged a conversation partner to be human, but the partner was ChatGPT GPT-4o relayed through a proxy, so the author concludes that AIs cannot yet reliably evaluate whether they are talking to a human or another AI. The piece also notes that spam filters already run such tests constantly, and cites Sam Altman and Dario Amodei on when superintelligence might arrive. |
Comments No comments yetWhat caught your eye? Start the discussion. |
More from interconnected.org on Kiwi News |

