OpenAI研究员5年前被MIT教授否定的PPT，预言o1与o3核心思路
五年前被MIT教授否定的“开放式推理”PPT，如今在OpenAI o1、o3上得到印证，值得一看。
2020年，现任OpenAI推理模型研究者Giambattista Parascandolo在MIT求职面试时汇报用GPT做推理的设想，被评委批为“无稽之谈”。其核心概念“开放式推理”主张模型可增加计算时间修正答案，正是如今o1、o3的推理时扩展思路；他还提出以语言为推理载体，类似今天的思维链。他本人将这段经历写进个人主页。
正文摘录
- title: A PPT Dismissed as "Nonsense" by MIT Professors 5 Years Ago Predicted the Core Ideas of OpenAI o1 and o3 - sourcecompany: 机器之心 - bodymarkdown: .jpg) Editor: 杨文 The AI circle never lacks gossip. This time, the protagonist is Giambattista Parascandolo, a key researcher on OpenAI's reasoning model team.  In 2020, he went to MIT for a faculty position interview and gave a talk about reasoning with GPT. In the end, most professors on the search committee dismissed this direction as "nonsense." ![图片](…