940: In Case You Missed It in October 2025

Episode Summary

Executive Summary: This episode recaps October highlights on AI’s rapid skills shift, practical optimization in enterprise operations, ethics in technology, new model architectures beyond transformers, and an AI startup’s unconventional go-to-market. The common thread is that AI is reshaping work from model-building to orchestration, while also demanding clearer ethical guardrails and business value focus.

Main Topics: AI is forcing continual professional rewiring: Seamus McGovern argues that the accelerating pace of AI means workers must get comfortable with constant skill turnover and learn to collaborate with AI as a partner rather than fear replacement. Optimization as a practical enterprise advantage: Real-world cases from Toyota and Total Wine show how mathematical optimization, combined with LLM interfaces, helps planners and data scientists make scenario-based decisions in complex supply-chain settings. Ethics and trust in technology: Stephanie Hare discusses why technologists may need an ethos similar to the Hippocratic Oath, emphasizing 'first do no harm' and balancing benefits against harms without relying solely on regulation. Beyond transformers: concatenation and sparse activation in BDH: Adrian Kosovsky explains how the BDH architecture can concatenate separately trained models and offers efficiency benefits, with future promise especially in reasoning models and very large-context enterprise use cases. Unframe’s unconventional AI business model: Larissa Schneider describes a 'managed AI delivery platform' that blends product and services, charges only when customers see value, and uses ROI/KPI-driven proof-of-value engagements.

Key Arguments: AI-related skills are turning over rapidly, so career resilience now depends on comfort with continuous change and rewiring rather than static expertise. AI should be viewed less as a tool and more as a collaborative partner that can augment or automate tasks, creating new work that was previously impossible. In white-collar work, the future is moving from building models from scratch to orchestrating, supervising, evaluating, and managing AI workflows. Optimization becomes far more accessible when planners and domain experts can interact with models through natural-language interfaces rather than coding directly. Technologists should internalize ethical principles akin to medicine's 'first do no harm' to maximize benefits and minimize harms in product design. BDH’s sparse activation and modularity enable model concatenation and may support more efficient reasoning models and massive-context applications. Unframe’s success suggests AI offerings may win by tightly coupling product, services, and measurable business impact rather than traditional SaaS-only packaging.

Data Points: Episode number: 940 - Super Data Science Podcast 'In Case You Missed It' October episode Past episode referenced: 933 - Interview with Seamus McGovern on AI skill rewiring Skills turnover in AI-exposed jobs: 30-something percent every ~3 years - McGovern citing a study on AI-exposed industries/jobs Compute growth: doubling every 6 months - McGovern describing accelerating AI infrastructure change Past episode referenced: 935 - Interview with Dr. Stephanie Hare on AI ethics Past episode referenced: 929 - Interview with Dr. Adrian Kosovsky on BDH architecture Model scale: 1B parameters - BDH demonstrations compared with GPT-2-scale models Context size example: 1 million pages = 1 billion tokens - Kosovsky describing enterprise reasoning-model ingestion potential Company raise: $50 million - Unframe’s total venture capital raised so far Launch timing: March last year - Unframe was started roughly in March of the prior year Seed round timing: Seed then A round in March this year - Unframe funding timeline Pricing model: t-shirt sizes: small, medium, large, extra-large - Unframe pricing by solution per year

Pivotal Quotes: "You have to get comfortable with the speed of change." — Seamus McGovern: Advice on adapting professional skills to rapid AI advances "First do no harm." — Dr. Stephanie Hare: Core ethical principle she suggests technologists should adopt "We build an AI platform that is made up of hundreds of different building blocks." — Larissa Schneider: Explaining Unframe’s modular managed AI delivery platform

Implications: Listeners should expect faster skill churn, greater demand for AI orchestration and evaluation roles, stronger pressure for ethical design, and more business models tied directly to measurable AI value rather than generic software.