Meta has introduced a new AI training technique called Thought Preference Optimization (TPO) to enhance how machines process information and respond to queries. TPO encourages language models to engage in internal reflection before providing answers, resulting in more nuanced and human-like responses. Unlike traditional methods, TPO allows AI to think independently in a single step, leading to more creative problem-solving abilities. By drawing inspiration from cognitive science, Meta aims to develop AI that understands complex reasoning processes. The TPO approach, tested against industry benchmarks, has shown promising results in improving AI performance on challenging tasks. Additionally, Meta's research on System 2 distillation aims to combine fast, intuitive System 1 processing with slow, analytical System 2 thinking in AI models. This innovative approach could potentially lead to the development of more intelligent and efficient open-source AI models without the need for extensive new data. Read more AI-generated news on: https://app.chaingpt.org/news