CriticGPT, Living Robot Skin, and Oyster-Inspired Concrete – Perplexity

In this episode of Discover Daily, we explore OpenAI's CriticGPT, an AI model designed to identify errors in code generated by ChatGPT. CriticGPT catches 85% of errors compared to 25% by humans and its feedback was preferred over human critiques in 63% of cases. OpenAI plans to integrate CriticGPT into their AI training pipeline, potentially leading to more reliable AI models.Next, we discuss a groundbreaking humanoid robot face created by scientists at the University of Tokyo and Harvard, co...