AI researcher's 5-year PhD retrospective: dead ends, multimodal detours, and a postdoc at Columbia
kchonyc · x · 2026-09-07
NYU PhD graduate dmadaan shares a long-form reflection after finishing a 5-year PhD under Kyunghyun Cho and starting a postdoc at Columbia DBMI. Key takeaways: no academic family background — he got in via FOR.ai (now Cohere Labs) and a fully funded KAIST MS; knowing why you pursue a PhD shapes your projects, patience, and advisor choice; his first project dead-ended, teaching him when to stop; and a year-long detour answering "what even is multimodal learning?" seeded his later work. A candid read for current and prospective PhD students.
More from Companies & People
- ByteDance preps real-time spatial video generation model, Zhang Yiming personally overseeing launch — pstAsiatech · 2026-09-07
- Law schools restrict AI only for tasks constitutive of a legal education, scholar explains — hoofnagle · 2026-09-07
- Senior DeepMind AlphaGo veteran leaves to build AI that 'properly reasons', skeptical of LLMs — ThoreG · 2026-09-07
- ELLIS PhD Recruiting Opens for Interpretability, Safety, AI for Science and Multi-Agent Research — boknilev · 2026-09-07
- AI economy hits $229B annualized revenue, up 3.5x in one year — BenBajarin · 2026-09-07
- Palantir revisited: 110% stock gain despite a GTM team the company seems to despise — thedealdirector · 2026-09-07