AI Innovators: High School Edition

Sathvik Subramanian

AI Innovators: High School Edition is a podcast that dives into the world of artificial intelligence through the voices of entrepreneurs, experts, and everyday professionals. In each episode, host Sathvik Subramanian explores how AI is transforming industries, empowering startups, and reshaping careers. From cutting-edge innovations to real-world applications, the podcast offers a deep look at the people and technologies driving the AI revolution. Tune in for insightful conversations and a collective exploration of AI’s impact on the future.

Episodes

  1. Aug 31

    The first generation not to get smarter. Dr. Tracey Tokuhama-Espinosa (Harvard): Learning Vs AI.

    This is the third episode in a series about how humans learn and how machines learn. Sumitra Ganesh (J.P. Morgan) told me small models are the most underhyped idea in AI because the brain runs on twenty watts. Anish Mahishi (NVIDIA) told me reasoning models finally work because they think one internal token at a time, the way humans do. Both answers pointed at the human brain. Dr. Tracey Tokuhama-Espinosa from Harvard has spent her career studying it. Dr. Tracey Tokuhama-Espinosa teaches the Neuroscience of Learning at Harvard and has spent three decades studying how humans actually learn. In this conversation she explains neuroconstructivism, the Five Pillars of the Mind, why cognitive offloading via ChatGPT literally removes the scaffolding your brain needs for higher-order thinking, why yours is the first generation not to get smarter, and the one non-technical piece of advice she gave a fifteen-year-old host. Guest Dr. Tracey Tokuhama-Espinosa. Google ScholarLinkedInNeuroscience of Learning course at Harvard ExtensionConexiones: The Learning Sciences Platform Books mentioned or referencedFive Pillars of the Mind: Redesigning Education to Suit the Brain (W. W. Norton, 2019)Mind, Brain, and Education Science (W. W. Norton, 2010)Bringing the Neuroscience of Learning to Online Teaching (Teachers College Press)Concepts and research referenced in the conversation Flynn effect (the rise, and recent decline, of average IQ)The IKEA effect (Norton, Mochon, Ariely)Neuroconstructivism (Karmiloff-Smith / Sirois / Westermann)Adolescent Brain Cognitive Development (ABCD) StudyThe Harvard Study of Adult Development (nearly a century long)Host Wisen AI. Sathvik's startup, creating training data by paying senior citizens for their tacit knowledgeNewark AcademyContact If you're a high schooler building something with AI, especially at the edge of AI and learning, send it to Sathvik: sathvik@wisenai.com

  2. Jul 27

    The AI you use was built twice - Anish Mahishi (NVIDIA NeMo-RL) on post-training

    Anish Mahishi is an engineer on the NeMo-RL team at NVIDIA. He works on the post-training systems that turn raw pretrained models into the AI you actually use — ChatGPT, Claude, and every real product built on top of a language model. Before NVIDIA he was a data scientist at Fractal, a Master's student at NYU (with medical MRI research at NYU Langone), an intern on the PyTorch team at Meta, and a machine learning engineer at Bloomberg for three years. In this episode we cover: How he got into AI (an Andrew Ng Coursera course in his second year of college)The full pipeline from you hitting enter in ChatGPT to the first word showing upWhat post-training actually is, and what the NeMo-RL team buildsRLHF vs. DPO vs. GRPO — the alphabet soup, translatedWhat DeepSeek R1 actually changed about how these systems runWhy fully autonomous agents are overhyped and evaluations are underhypedWhy medical AI is the hardest domain to get rightOn-device AI, and the one non-technical skill that mattered more than he expectedGuest Anish Mahishi — Google ScholarAnish Mahishi — LinkedInAnish's research On the feasibility of machine learning augmented magnetic resonance for point-of-care identification of disease (arXiv 2301.11962, 2023) Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks (arXiv 2605.01417, May 2026) Medmarks on GitHubConcepts and tools we discussed Andrew Ng's Machine Learning course on Coursera NVIDIA NeMo-RL on GitHub Training language models to follow instructions with human feedback (InstructGPT / RLHF paper)Direct Preference Optimization (DPO paper)DeepSeekMath — GRPO paperDeepSeek-R1 — the reasoning-model recipe we discussedPyTorch Wisen AI — Sathvik's startup, bridging older-adult experience and modern AINewark AcademyContact If you're a high schooler building something with AI, send it to Sathvik: sathvik@wisenai.com

  3. Jul 6

    AI Agents - Hype vs Real Research. A Conversation with Dr. Sumitra Ganesh of JP Morgan

    Dr. Sumitra Ganesh heads AI Research in JP Morgan. Enjoy a fascinating conversation about her research. ⁠Dr. Sumitra Ganesh — bio page at J.P. Morgan AI Research⁠⁠Sumitra Ganesh on Google Scholar⁠ — full publication list⁠J.P. Morgan AI Research⁠1. The 2019 dealer-market paper ⁠Reinforcement Learning for Market Making in a Multi-agent Dealer Market (Ganesh et al., NeurIPS 2019)⁠2. Phantom ⁠Phantom on GitHub⁠⁠Phantom — A RL-driven Multi-Agent Framework to Model Complex Systems (Ardon, Vann, Garg, Spooner, Ganesh — AAMAS 2023)⁠⁠Phantom documentation⁠3. The 2021 paper ⁠Towards a fully RL-Based Market Simulator (Ardon, Vadori, Spooner, Xu, Vann, Ganesh — ICAIF 2021)⁠4. The 2020 NeurIPS paper Dr. Sumitra referenced ⁠Calibration of Shared Equilibria in General Sum Partially Observable Markov Games (Vadori, Ganesh, Reddy, Veloso — NeurIPS 2020)⁠5. The 2022 paper that extended the framework to OTC markets ⁠Towards Multi-Agent Reinforcement Learning driven Over-The-Counter Market Simulations (Vadori, Ardon, Ganesh et al. — 2022; Mathematical Finance, 2024)⁠6. The 2025 Annual Reviews piece ⁠Generative AI Agents for Knowledge Work Augmentation in Finance (Annual Reviews of Control, Robotics, and Autonomous Systems, Vol. 8, May 2025)⁠The paper she said most influenced her thinking recently ⁠Artificial intelligence model can detect Parkinson's from breathing patterns (MIT News, Aug 2022)⁠⁠Dina Katabi — MIT EECS faculty page⁠The tool she said surprised her most in the last two years - Claude Code WIsenAI.com - Sathvik's start up. He reads everything - send your comments to sathvik@wisenai.com

About

AI Innovators: High School Edition is a podcast that dives into the world of artificial intelligence through the voices of entrepreneurs, experts, and everyday professionals. In each episode, host Sathvik Subramanian explores how AI is transforming industries, empowering startups, and reshaping careers. From cutting-edge innovations to real-world applications, the podcast offers a deep look at the people and technologies driving the AI revolution. Tune in for insightful conversations and a collective exploration of AI’s impact on the future.