Parascandolo's 2020 MIT interview presentation proposed three key ideas: open-ended reasoning (investing more compute for harder problems), language as reasoning carrier, and system self-modification—concepts now central to modern reasoning AI. The researcher, trained under Bernhard Schölkopf at ETH Zurich with internships at Google X and DeepMind, joined OpenAI's reinforcement learning team and later led work on the o1 and o3 projects.