641: Data Science Trends for 2023

Episode Summary

Executive Summary: John Crohn and Sadie St. Lawrence reviewed 2022 data-science predictions, then forecast 2023 around data as a product, multimodal generative AI, decentralized enterprise data, AI policy, and sustainability. The discussion highlighted ChatGPT’s explosive adoption, the shift from hype-driven blockchain/NFT use cases to practical audit-trail applications, and the growing need for data literacy and governance as AI becomes consumer-facing and API-accessible.

Main Topics: Review of 2022 prediction accuracy: The hosts assessed prior predictions on AutoML, GANs, scalable ML operations, blockchain data utility, and data literacy, noting most were directionally correct while GANs were already fading and a major missed trend was generative AI / ChatGPT. Data as a product: Sadie’s main 2023 theme: organizations and consumers will increasingly interact with data through products, especially large-language-model interfaces, and enterprises will treat unique internal data as monetizable assets. Multimodal generative models: They predicted continued breakthroughs in text-to-image, text-to-video, and other multimodal systems, with compelling video generation as a likely 2023 milestone. Data mesh and decentralized enterprise data: The conversation framed data mesh as a business-and-operations shift, not just a technical architecture, requiring self-service access, standards, privacy controls, and stronger organizational readiness. AI policy, privacy, and trust: They discussed the US AI Bill of Rights, rising regulation, lawsuits, algorithmic discrimination concerns, and the need for legal as well as technical governance as AI reaches consumers. Sustainability and accelerated compute: The episode addressed the energy costs of larger models, the need for more compute per watt, and the importance of sustainable infrastructure and policy for data centers and AI workloads. Personal and professional updates: Sadie shared her work on Women in Data, a world tour, blockchain-based education credentials, mentorship, and a future book; the episode also included CrossFit predictions and a book recommendation.

Key Arguments: AutoML’s practical value has expanded, but the broader societal risks of automation are increasingly visible in consumer-facing tools that may embed bias. GANs were an overhyped 2022 trend; generative AI shifted attention toward foundation models with broader utility and adoption. Data-science teams should design for scalability, monitoring, traceability, and iteration from the outset because model lifecycle automation is now central to production AI. Blockchain’s strongest future may be in auditability, traceability, and non-financial use cases rather than speculative crypto/NFT hype. NFTs are being rebranded as digital collectibles, suggesting the underlying technology may persist while the marketing language changes. Data literacy remains a major bottleneck; widespread AI adoption will depend on educating business users, not just building better models. Data mesh adoption will be limited unless organizations also change people, process, and governance; technology alone cannot deliver the model. AI policy is moving from abstract ethics to enforceable frameworks and case law, making legal oversight a core requirement for model deployment. Sustainable AI will require both better hardware/compute efficiency and cleaner energy sources; organizations should ask about the sustainability of their data centers. Foundation models are changing the innovation model: many teams will consume APIs and fine-tune instead of training foundational models from scratch. Large language models are powerful but can hallucinate confidently incorrect answers, making verification and critical thinking essential in education and work. Healthcare AI and AI safety were identified as especially promising and impactful career areas for new entrants.

Data Points: 2022 prediction score: 4 out of 5 - John assessed last year’s predictions as largely correct, with one major miss around generative AI. Data literacy share of global workforce: about 21% - Sadie cited this as the low baseline for data literacy worldwide. Women in Data community size: over 20,000 women across 17 countries - Sadie described the reach of her WomenInData.org community. Students reached by Sadie’s content: over 350,000 students - Introduced as part of her background in data science and machine learning education. ChatGPT time to 1 million users: 5 days - Sadie used this as evidence of unusually fast consumer adoption. Comparable platforms to 1 million users: about 2.5 months - She contrasted ChatGPT’s growth with earlier major platforms. Meta/Facebook value change: lost 80% of its value - John referenced Meta as an example of hype and valuation correction. CrossFit Open participation fee: $20 - John explained that anyone can join the CrossFit Open by paying a small fee.

Pivotal Quotes: "Data as a product" — Sadie St. Lawrence: Sadie’s headline theme for 2023, encompassing both consumer-facing AI and enterprise data monetization. "ChatGPT has what I call broke the internet" — Sadie St. Lawrence: Used to underscore how foundational models moved AI into mainstream consumer use. "What problem are you solving?" — Sadie St. Lawrence: Her go-to anti-hype question for startups and buzzword-heavy pitches.

Implications: The episode argues that 2023 will be defined by consumer AI adoption, enterprise data commercialization, and stronger governance. Teams that combine data literacy, sustainable infrastructure, and trustworthy deployment will be best positioned to benefit.