Neuroscientist built Socrates, an AI that refuses to give answers

Theoretical neuroscientist Vivienne Ming built Socrates, a model deliberately scoring zero on every benchmark that offers only context and questions, never answers. In her prediction experiment, the share of users who shifted into "cyborg" mode rose from 21% to 30%.

• Socrates scores zero on every benchmark and withholds answers
• Share of "cyborg" users rose from 21% to 30%
• 60–70% of participants just copy-pasted chatbot answers, another 25% sought validation
• Cyborg teams using a small open-source model beat four standalone models on accuracy