翁家翌：OpenAI，GPT，强化学习，Infra，后训练，天授，tuixue，开源，CMU，清华｜WhynotTV Podcast
FreeInside OpenAI: RLHF, Post-Training, and Open Source with Weng Jiayi
FreeFree tier
About 翁家翌：OpenAI，GPT，强化学习，Infra，后训练，天授，tuixue，开源，CMU，清华｜WhynotTV Podcast
This YouTube-style Bilibili podcast episode features Weng Jiayi (翁家翌), a key contributor to OpenAI's GPT series (GPT-3.5, GPT-4, GPT-5) who joined in 2022. The conversation covers his childhood, education at Tsinghua University and CMU, his work on reinforcement learning, post-training, and infrastructure at OpenAI, and his open-source projects including the Tianshou RL framework and the tuixue visa tracking system. It explores his philosophy of prioritizing impact over recognition and offers an insider perspective on OpenAI's culture, the ChatGPT launch, and the path to AGI.
Key Features
Interview with OpenAI researcher Weng Jiayi on GPT-3.5/4/5 development
In-depth discussion on reinforcement learning and post-training (RLHF)
Stories behind open-source projects: Tianshou RL framework and tuixue visa tracker
Insider perspective on OpenAI's infrastructure and team dynamics
Coverage of his academic journey from Tsinghua to CMU to OpenAI
Philosophical views on impact, open source as charity, and career choices
Pros & Cons
Pros
- First-hand account from a core OpenAI contributor on GPT model evolution
- Detailed explanation of RLHF and post-training for non-experts
- Covers both technical depth and personal career narrative
- Highlights the importance of open-source contribution and impact
- Includes practical insights into industrial-scale RL infrastructure
Cons
- Video is over two hours long; requires time investment
- Content primarily in Chinese; may limit accessibility
- Not a hands-on tutorial; focused on discussion and reflection
- Some topics assume familiarity with AI and reinforcement learning concepts
Best For
Learning about reinforcement learning and post-training in large language modelsUnderstanding the history and challenges of RLHF at OpenAIGaining motivation from an open-source advocate's career storyExploring the culture and internal workings of OpenAIFor students and researchers interested in AI, NLP, and infrastructure
FAQ
Who is Weng Jiayi?
Weng Jiayi is an OpenAI researcher who joined in 2022 and contributed to GPT-3.5, GPT-4, and GPT-5, focusing on reinforcement learning, post-training, and infrastructure. He also created open-source projects such as the Tianshou RL framework and the tuixue visa tracking system.
What topics are covered in this podcast?
The podcast covers Weng Jiayi's childhood, high school programming contests, undergraduate studies at Tsinghua, research in reinforcement learning, internships at Yoshua Bengio's lab, his master's at CMU, joining OpenAI, the development of RLHF, ChatGPT's launch, and reflections on open source and impact.
What is Tianshou?
Tianshou is an open-source reinforcement learning framework created by Weng Jiayi during his undergraduate studies at Tsinghua. It aims to provide a modular and efficient toolkit for RL research.
What is tuixue?
Tuixue is an online visa tracking system developed by Weng Jiayi to help students and researchers monitor their visa application status. It became widely used in the Chinese overseas community.