Andrew Dai

Andrew Dai spent 12 years as a Research Scientist at Google Brain and DeepMind, working on foundational techniques for modern language models. He authored a 2015 paper that OpenAI later cited as the original recipe for ChatGPT and held core leadership roles on Gemini, GLaM, and PaLM 2. His 2026 AI Engineer World’s Fair biography reported more than 67,000 citations to his published research and identified him as co-founder and CEO of Elorian, a company building AI systems that understand visual information and reason in ways modeled on human thinking. The biography described Elorian’s launch with $55 million at a $300 million valuation, backed by Menlo Ventures, Altimeter, Striker Venture Partners, NVIDIA, and Jeff Dean.

In his conference presentation, Dai examines how pattern recognition can substitute for careful visual reasoning, producing errors in counting, spatial relationships, and tracking actions across video. He presents Elorian’s approach: specialized visual reasoning data, a synthetic data flywheel combining evaluations, agents, supervised fine-tuning, and reinforcement learning, improvements to transformer architectures, and visual chain of thought. One example uses bounding boxes to identify hotels before selecting the red ones. He proposes applications in robotic scene understanding, connecting construction footage to written safety policies, and extracting geometric constraints from designs for programmatic or simulation validation.