CAPTCHAs can still detect AI agents
A new research paper proposes a "Process Turing Test" to distinguish AI agents from humans, moving beyond simply evaluating output to analyzing the cognitive processes involved. The study found that while advanced models like GPT and Claude perform similarly to humans on CAPTCHA tasks, their underlying methods differ significantly. Surprisingly, smaller open-source models like Qwen and Centaur exhibited more human-like cognitive processes than their larger, frontier counterparts. AI
IMPACT AI detection methods are evolving beyond simple output analysis, potentially impacting bot mitigation strategies and the understanding of AI 'humanness'.