640: What I Learned in 2022

Episode Summary

Executive Summary: The episode is a year-end reflection on 2022, highlighting major growth for the Super Data Science Podcast, the expansion of its team, and five personal lessons from the host: accepting limits on time, marveling at larger AI models, stepping back from the 24-hour news cycle, valuing in-person work, and using nutrition logging for health and consistency. It ends with gratitude, a call for balance, and optimism for 2023.

Main Topics: Podcast and team growth in 2022: The host reviews the show’s strongest year yet, citing major audience growth, stronger production quality, and the addition of Serge Masis, Dr. Zara Karche, and Natalie Zzaiski to support research, writing, and operations. Lesson 1: Limits on time and attention: The host explains he could not sustain every project at once, especially YouTube/Udemy math tutorials and the planned book, because podcast and company growth consumed available capacity. Lesson 2: Scaling AI models unlocks new capabilities: He argues that orders-of-magnitude larger models produced surprising, human-like generative abilities in 2022, citing GPT-3, DALL·E 2, ChatGPT, and Cicero as evidence. Lesson 3: The 24-hour news cycle is draining: The host reflects on how pandemic-era news consumption became a habit that feels exhausting and unsatisfying, and says he wants to return to slower, deeper reading and more leisure reading. Lesson 4: In-person work is more enjoyable: While acknowledging remote work works, he says face-to-face collaboration, off-sites, and conferences increased fun, motivation, and social energy compared with Zoom-only interaction. Lesson 5: Nutrition logging supports health and freedom: He describes consistent macro tracking, gym work, mobility training, and team CrossFit as helping him lose body fat, improve performance, and feel less mentally burdened by eating decisions. Gratitude, balance, and outlook for 2023: The episode closes with thanks to the podcast team and Nebula, an admission that work remains over-dominant, and a promise of more ambitious AI/data science content in the coming year.

Key Arguments: The podcast’s growth and quality improved because of added team roles and a stronger production pipeline. The host cannot simultaneously maintain the podcast, company-building, YouTube, Udemy, and a book at the same pace. Larger AI models are producing emergent, surprisingly human-like capabilities that were not anticipated by their creators. Constant news consumption worsens mood and is less satisfying than slower, curated reading. In-person interaction improves morale and makes work more enjoyable, even if remote work remains effective. Consistent nutrition tracking, exercise, and mobility work produce sustainable body-composition and performance gains. The host’s life and work remain highly productive but still out of balance, with insufficient leisure and personal time.

Data Points: Podcast download growth: 90% - Most recent quarter versus the same quarter a year earlier Guest episode listens: At least 35,000 listens - Conservative estimate for individual guest episodes Annual episode output: 104 episodes - The podcast publishes 104 episodes per year Udemy students: Over 100,000 students - Math for machine learning course enrollment Body fat percentage peak: Above 17% - Peaked in November 2020 before the nutrition and training changes Body fat percentage current: Below 11% - Reported after about two years of consistent tracking and training Training frequency: 5 to 6 times a week - Consistent gym attendance over the last year Mobility practice frequency: At least 6 days a week - Use of the Pliability app for stretching and mobility Book reading since pandemic start: 1 book - The host says he has barely read for leisure since COVID, except for 4,000 Weeks Podcast episode references: 539, 559, 569, 589, 606, 618, 628, 630, 632, 633, 634, 641 - Episodes mentioned as examples of guests, recordings, or follow-up discussions Cicero performance: 90th percentile or higher - MetaAI’s Diplomacy-playing system performance

Pivotal Quotes: "I can't do everything at once." — Host: Core first lesson about prioritization and accepting limits "Orders of magnitude more parameters produce unbelievable AI models." — Host: Second lesson about the rapid advance of generative AI "Working in person is way more fun." — Host: Fourth lesson emphasizing the emotional benefits of face-to-face collaboration

Implications: Listeners are encouraged to focus on priorities, expect rapid AI progress, reduce media overload, seek more in-person connection, and use consistent tracking/habits for health. The industry takeaway is that 2023 could bring even bigger AI breakthroughs and stronger podcast/community engagement.