After 10+ years of teaching @stanfordnlp 's Spoken Language Processing course, I’ve seen some amazing research projects, but I was still shocked by @AliAnsariMicro1 's project check-in. He described thousands of candidates in 30-minute conversation sessions with an AI skills interviewer. Experimenting to improve LLM-based conversations with real humans at this scale is a complex, fascinating research challenge. Ali and I kept in touch as @micro1_ai continued its meteoric rise in providing human expert data + feedback to the world’s leading AI efforts. Modern AI models enable experts across broad domains to inject detailed knowledge, reasoning chains, and multimodal context to expand AI capabilities and improve correctness. I’m thrilled to join @micro1_ai as VP of AI! Our work with partners is inventing the next generation of data-centric deep learning. Modern AI models need specialized training data and feedback; longstanding challenges like data quality, diversity, and bias manifest differently with RL / reward post-training mechanisms, multimodal foundation models, and physical AI / robotics. @micro1_ai is solving these exciting challenges, and there is work to do! #DeepLearning #DataCentric