關於
Mark Chen is the Chief Research Officer at OpenAI, where he oversees the company's research priorities and direction toward artificial general intelligence. Before joining OpenAI in 2018, he worked as a quantitative trader at Jane Street Capital and Integral Technology. A graduate of MIT with a degree in Mathematics with Computer Science, Chen has been instrumental in developing landmark AI systems including DALL-E, Codex, and GPT-4, and played a key role in building OpenAI's o1 and o3 reasoning models.
主要貢獻
- Co-led DALL·E, one of the first systems to make text-to-image generation legible to the public
- Helped build Codex, connecting language-model research to practical code generation and developer tools
- Contributed to GPT-4-era multimodal and frontier-model research before moving into top research leadership
- Worked on OpenAI's reasoning-model line, including the o-series shift from next-token fluency toward deliberative problem solving
- Became OpenAI's Chief Research Officer, putting him in charge of research direction at one of the most consequential closed AI labs
- His low public profile contrasts with the power of the role, raising the broader question of how much frontier-AI research accountability depends on company insiders
影片與訪談
思想連結
Ilya Sutskever
曾經合作共同創辦人・Safe Superintelligence Inc.
Chen 是〈Evaluating Large Language Models Trained on Code〉（2021）的第一作者——那篇把語言模型變成程式設計工具的 Codex 論文；Sutskever 的名字在五十八位作者的末端，首席科學家署名的位置。這篇論文標記了「什麼算證據」的轉向：從模型預測得多準，轉向模型實際被測出能做到什麼。Chen 後來接下了 Sutskever 曾據守的研究位置，成為 OpenAI 的首席研究長，卻幾乎沒有隨之而來的公共能見度。
arxiv.org
Terence Tao
曾經對談Professor of Mathematics, UCLA
收錄於本站的這場 IPAM 爐邊對談，已是兩人三年來的第三次；主持人開場時，重提了陶哲軒早先對這些模型之於數學的評價：大概相當於一個很沒效率的研究生。這正是這個系列的價值所在——Chen 來自一間打造推理模型的實驗室，陶哲軒則來自一種「證明要嘛被驗證、要嘛什麼都不是」的實作，而每年重來一次，把一場分歧變成了一次量測。陶哲軒的興趣格外具體：他實際使用這些系統，並如實回報它們錯在哪裡。
youtube.com
