Sep 14, 2023 · 38m · mad
From Xbox to Databricks: Carly Taylor’s Rebel Path in Data Science & Gaming AI
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The MAD Podcast, host Matt Turck interviews Carly Taylor, founder of Rebel Data Science, about her journey from computational chemistry to Activision, AI applications in gaming and cheat detection, model observability, and marketing strategies for tech startups.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 26.7% of the talking time here. How this is scored →
Matt as informed peer 2.9 Guest teaching 3.9 Guest disagreement 0.1 Matt pushing back 0.4
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 28:15 Gentle pushback on X/Twitter vs LinkedIn value
Carly counters the conventional tech view that Twitter/X is superior to LinkedIn for AI startups, arguing that LinkedIn offers a far deeper bench of decision-makers.
Hardest push from Matt ▶ 12:31 Testing the underlying economic driver of GenAI NPCsMatt pushes past general enthusiasm to clarify whether the true driver of GenAI NPCs is developer cost avoidance rather than novel technical capability.
Biggest teaching moment ▶ 8:40 Explaining adversarial machine learning dynamicsCarly educates Matt on how security machine learning flips traditional ML methodology by focusing on outliers and accounting for how model decisions alter human behavior.
Matt holds his own ▶ 19:44 Matt articulates the venture capital reality of distributionMatt asserts strong domain knowledge as a VC, pointing out that technical superiority leads to failure in obscurity without dedicated distribution strategies.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome and Carly Taylor's Career Path from Chemistry to Gaming | 1 | 2 | 0 | 0 | Matt introduces Carly warmly and asks for a brief background on her transition into gaming and data science. Carly explains her unconventional path from computational chemistry into data science and Activision. The atmosphere is highly agreeable and conversational. | |
| The Intersection of AI, Machine Learning, and Video Games | 2 | 3 | 0 | 0 | Matt prompts Carly on how machine learning applies to large gaming franchises like Call of Duty within corporate IP constraints. Carly details how ML has expanded from standard graphics rendering into audio synchronization and workflow acceleration. | |
| AI in Security, Anomaly Detection, and Adversarial Machine Learning | 3 | 5 | 0 | 1 | Carly educates Matt on the mechanics of anomaly detection and adversarial machine learning using a credit card fraud analogy. Matt synthesizes her point by reframing it as a broader societal dynamic around ML adoption. | |
| Generative AI Applications in Non-Player Characters and Custom Gameplay | 4 | 4 | 0 | 1 | Matt probes into Generative AI applications in gaming, specifically asking if cost-efficiency is what makes AI non-player character customization viable. Carly agrees with his framing and discusses dynamic, unscripted quest creation. | |
| Data Science Team Structures and Representation in Gaming Studios | 2 | 4 | 0 | 0 | Matt asks how gaming companies structure data science teams and how Carly experienced representation in the field. Carly breaks down the trade-offs between centralized centers of excellence and embedded domain models. | |
| Building Rebel Data Science, Branding, and Community Distribution Strategies | 4 | 5 | 1 | 1 | Matt offers VC perspective on how technical startups fail without distribution strategies and asks Carly for advice. Carly provides an extended masterclass on content repurposing, community building, and why LinkedIn often outperforms X for tech founders. | |
| Technical Tooling, Open-Source ML, and Machine Learning Observability | 4 | 4 | 0 | 0 | Matt demonstrates technical familiarity with ML tooling by asking Carly to break down specific components of observability like data lineage and drift. Carly shares insights on label drift prevention in production environments. |
Statements from this episode (10)
Taylor: Security machine learning must hyper-focus on outliers instead of discarding them
“There's a tenant of machine learning where like you just throw out the outliers because they're going to mess up your distribution and you kind of don't want to deal with them. For security, what you do is you find the outliers and you hyper focus on them beca…”
Taylor: Deploying machine learning models fundamentally alters the targeted adversarial problems
“And it's something that I think traditional machine learning hasn't really been agile enough to deal with. Right. Like the act of doing machine learning is fundamentally changing the problem you're trying to solve.”
Taylor: GenAI in games introduces new capabilities rather than replacing human jobs
“It's no one's Full-time job to make an NPC, like say your name. It's just a functionality we've never had. So you're not really replacing anything people were doing. You're just making something that didn't exist exist.”
Carly Taylor: Centralized data teams lose domain depth, embedded teams lose standards
“As soon as you centralize something, you will inevitably lose the deep expertise you can get from embedding, but as soon as you embed everyone, you lose that, like, you know, Center of excellence where everyone comes together and you set standards for your dat…”
Carly Taylor: Gaming data teams have better diversity than core programming roles
“I see more representation in, in data teams than I do for something, let's say like the, I don't know, hardware level programming. You know, which has just historically been, like, a lot of these, like, deep nitty-gritty computer science fields have been, like…”
Taylor: Community building provides more startup growth leverage than marginal technical improvements
“Sometimes you can over index on the technical and getting maybe, you know, your platform to be five percent more fast or supported on one more browser might not 10 X you the way that spending that time building a community would.”
Taylor: Data science and AI startups should prioritize LinkedIn over Twitter
“Twitter might get you more impressions, I think. And you might get your message out wider. So I don't think it has to be an either or, but if I was going to spend a lot of my time, especially if I was a data science or AI startup, I would learn where my audien…”
Taylor: Model observability is the most important focus for data scientists
“I harp on observability a lot because I think it's, like, probably the most important thing a data scientist can focus on”
Taylor: Startups without ready data must prioritize hiring a cloud data engineer
“If you don't have your data ready yet, you can't skip the data engineering piece of this, and I'd say you probably need someone who's going to be like your cloud data engineer. Like you just have to have those basics covered”
Taylor: Healthcare lags eight years in tech, creating massive upside for ML
“I think that health care is historically like eight years behind everyone else, but there's also a very massive upside there right now for people who are going to be using machine learning and data science to solve problems.”