The Global Institute of Data Science

04/28/2026

Generative Narratives is my new Substack

The Quiet Disappearance of the Human Translator

04/21/2026

For four decades, quant finance lived inside three variables. Fama-French market, size, and value. It worked . . . Marketing did the same thing — Nielsen 3C. Logistics did the same thing — Time-Volume-Value. Three industries, three frameworks, one shape. None of them coordinated.

At Future Alpha 2026, Dr. Bryan Kelly of Yale SOM and AQR opened the conference with the argument that this entire convention is ending.

His published research: take the same three factors, expand them into thousands of derived features inside a 10,000-parameter neural network, and out-of-sample performance triples.

Same inputs. No new data. Just more model.

The preference for simple models, in his words, is a learned bias — not an empirically justified rule.

Finance is going first. Your industry is next.

What are you seeing?

Why a 40-year-old rule in Machine Learning is quietly breaking across every industry.

03/10/2026

YouTube commands 13.4% of all U.S. television watch-time.

Netflix — spending $17 billion a year on content — commands 8.8%.

That gap isn't a content problem.

It's a data architecture decision made years before either company knew they were competing.

I spent two weeks mapping what Netflix, Paramount, and Warner Bros. actually built — and what they didn't.

The pattern repeats in financial services, manufacturing, and professional services.

One question worth sitting with:

In your industry, who controls the data feedback loop — and is it you?

——
Full analysis in first comment.

02/26/2026

Is there a data type you CANNOT do forecasting with?

Most people think forecasting = numbers in a spreadsheet.

But in 2026, we forecast with satellite images, call center audio, security camera video, social media engagement patterns, server logs, and sensor feeds.

The constraint is never the data type. It's whether your organization has invested in the pipeline to encode it.

I wrote a deep-dive on the full spectrum of forecastable data types — from structured to unstructured to the specialized modalities most companies are ignoring entirely.

The 85% AI project failure rate won't improve until organizations stop forecasting with one eye closed.

Full article on Substack

The question isn’t whether AI can forecast with your data. It’s why you’re still ignoring most of it.

02/17/2026

https://open.substack.com/pub/gidsco/p/i-spoke-about-ai-at-carnegie-mellon?utm_campaign=post-expanded-share&utm_medium=web

30 Entertainment Students Asked Better AI Questions Than Most Executives.

02/10/2026

𝐃𝐨 𝐲𝐨𝐮 𝐤𝐧𝐨𝐰 𝐭𝐡𝐞 𝐝𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐜𝐞 𝐛𝐞𝐭𝐰𝐞𝐞𝐧 𝐚𝐮𝐭𝐨𝐧𝐨𝐦𝐨𝐮𝐬 𝐚𝐧𝐝 𝐚𝐮𝐭𝐨𝐦𝐚𝐭𝐞𝐝?

Most executives don't. And right now, 1.5 million AI agents on a platform called 𝐌𝐨𝐥𝐭𝐛𝐨𝐨𝐤 are proving why that distinction matters along with 𝐎𝐩𝐞𝐧𝐂𝐥𝐚𝐰.

The headlines say it's the singularity. The technical reality? A sophisticated autocomplete running on a 30-minute timer.

This is the same cognitive trap that explains why 85% of AI projects fail: we evaluate the paint job and skip the structural inspection.

New video breaks it down in a few minutes.

02/03/2026

Speaking at 𝐂𝐚𝐫𝐧𝐞𝐠𝐢𝐞 𝐌𝐞𝐥𝐥𝐨𝐧'𝐬 𝐌𝐚𝐬𝐭𝐞𝐫 𝐨𝐟 𝐄𝐧𝐭𝐞𝐫𝐭𝐚𝐢𝐧𝐦𝐞𝐧𝐭 𝐈𝐧𝐝𝐮𝐬𝐭𝐫𝐲 𝐌𝐚𝐧𝐚𝐠𝐞𝐦𝐞𝐧𝐭 program on February 13th.

The topic: "𝘞𝘩𝘰 𝘖𝘸𝘯𝘴 𝘵𝘩𝘦 𝘋𝘢𝘵𝘢? 𝘊𝘦𝘭𝘦𝘣𝘳𝘪𝘵𝘺 𝘝𝘢𝘭𝘶𝘦, 𝘗𝘭𝘢𝘵𝘧𝘰𝘳𝘮 𝘌𝘤𝘰𝘯𝘰𝘮𝘪𝘤𝘴, 𝘢𝘯𝘥 𝘞𝘩𝘺 85% 𝘰𝘧 𝘈𝘐 𝘗𝘳𝘰𝘫𝘦𝘤𝘵𝘴 𝘍𝘢𝘪𝘭"

When fans engage with artists on Instagram, who captures the value — the talent, the platform, or the fan? And why do most enterprise AI projects still fail despite record investment?

These aren't technical problems. 𝐓𝐡𝐞𝐲'𝐫𝐞 𝐥𝐞𝐚𝐝𝐞𝐫𝐬𝐡𝐢𝐩 𝐩𝐫𝐨𝐛𝐥𝐞𝐦𝐬.

Looking forward to the conversation with Heinz College's entertainment management students in Los Angeles.

10/02/2025

𝐌𝐨𝐯𝐢𝐧𝐠 𝐁𝐞𝐲𝐨𝐧𝐝 𝐭𝐡𝐞 𝐂𝐨𝐝𝐢𝐧𝐠 𝐌𝐲𝐭𝐡—𝐏𝐫𝐚𝐜𝐭𝐢𝐜𝐚𝐥 𝐀𝐈 𝐆𝐮𝐢𝐝𝐚𝐧𝐜𝐞 𝐟𝐫𝐨𝐦 𝐭𝐡𝐞 𝐆𝐥𝐨𝐛𝐚𝐥 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞

For years, the tech industry pushed the idea that 𝐞𝐯𝐞𝐫𝐲𝐨𝐧𝐞 needed to learn manual coding to succeed in AI—regardless of background or business focus. But the reality of modern AI is changing fast, and the real opportunities lie elsewhere. Here’s what you need to know to transform your organization with AI today.

⚡ 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞 𝐭𝐡𝐞 𝐈𝐧𝐝𝐮𝐬𝐭𝐫𝐲 𝐁𝐢𝐚𝐬

⚡ 𝐀𝐈 𝐈𝐬 𝐌𝐨𝐫𝐞 𝐓𝐡𝐚𝐧 𝐂𝐨𝐝𝐢𝐧𝐠

Most of today's machine learning algorithms are decades old. Big Tech’s marketing campaigns, university bootcamps, and slick ads once insisted manual coding was essential for doctors, truck drivers, and every professional. But that myth started to fade rapidly with the rise of tools like ChatGPT and AI automation platforms.

💡 𝐊𝐧𝐨𝐰 𝐭𝐡𝐞 𝐃𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐜𝐞: 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞 ≠ 𝐂𝐨𝐦𝐩𝐮𝐭𝐞𝐫 𝐒𝐜𝐢𝐞𝐧𝐜𝐞

💡 𝐑𝐢𝐠𝐡𝐭 𝐒𝐤𝐢𝐥𝐥𝐬, 𝐑𝐢𝐠𝐡𝐭 𝐑𝐨𝐥𝐞𝐬
- Data science is not the same as computer science.
- Software engineering is not the same as machine learning engineering.
- A chief technology officer (CTO) is not a chief data scientist.

Success in AI comes from understanding these distinctions and building teams with the 𝐫𝐢𝐠𝐡𝐭 𝐞𝐱𝐩𝐞𝐫𝐭𝐢𝐬𝐞 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐫𝐢𝐠𝐡𝐭 𝐭𝐚𝐬𝐤.

🤖 𝐁𝐞𝐬𝐭 𝐏𝐫𝐚𝐜𝐭𝐢𝐜𝐞𝐬 𝐟𝐨𝐫 𝐄𝐧𝐝 𝐔𝐬𝐞𝐫𝐬 𝐚𝐧𝐝 𝐋𝐞𝐚𝐝𝐞𝐫𝐬

- 🤖 𝐋𝐞𝐯𝐞𝐫𝐚𝐠𝐞 𝐍𝐨-𝐂𝐨𝐝𝐞 𝐚𝐧𝐝 𝐀𝐮𝐭𝐨𝐦𝐚𝐭𝐢𝐨𝐧 𝐓𝐨𝐨𝐥𝐬
Empower business users and domain experts to drive AI adoption without forcing everyone to code.
- 🚦 𝐑𝐞𝐜𝐫𝐮𝐢𝐭 𝐟𝐨𝐫 𝐒𝐩𝐞𝐜𝐢𝐚𝐥𝐢𝐳𝐞𝐝 𝐃𝐚𝐭𝐚 𝐒𝐤𝐢𝐥𝐥𝐬:
Hire and train true data scientists—not just coders or general technologists.
- 📚 𝐀𝐥𝐢𝐠𝐧 𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐑𝐨𝐥𝐞𝐬 𝐰𝐢𝐭𝐡 𝐁𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐍𝐞𝐞𝐝𝐬:
Ensure CTOs, machine learning engineers, and software engineers are deployed where their expertise has the most business impact.

🎓 𝐓𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦 𝐰𝐢𝐭𝐡 𝐭𝐡𝐞 𝐆𝐥𝐨𝐛𝐚𝐥 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞

At GIDS, we help organizations break free from outdated thinking—guiding you to adopt AI in a way that matches your real needs, talent, and future strategy.
𝐏𝐫𝐚𝐜𝐭𝐢𝐜𝐚𝐥, 𝐩𝐨𝐰𝐞𝐫𝐟𝐮𝐥, 𝐚𝐧𝐝 𝐭𝐚𝐢𝐥𝐨𝐫𝐞𝐝 𝐟𝐨𝐫 𝐲𝐨𝐮𝐫 𝐩𝐚𝐭𝐡 𝐟𝐨𝐫𝐰𝐚𝐫𝐝.

𝐑𝐞𝐚𝐝𝐲 𝐭𝐨 𝐦𝐨𝐯𝐞 𝐛𝐞𝐲𝐨𝐧𝐝 𝐭𝐡𝐞 𝐡𝐲𝐩𝐞 𝐚𝐧𝐝 𝐦𝐚𝐤𝐞 𝐀𝐈 𝐰𝐨𝐫𝐤 𝐟𝐨𝐫 𝐩𝐞𝐨𝐩𝐥𝐞, 𝐧𝐨𝐭 𝐣𝐮𝐬𝐭 𝐩𝐫𝐨𝐠𝐫𝐚𝐦𝐦𝐞𝐫𝐬? 𝐏𝐚𝐫𝐭𝐧𝐞𝐫 𝐰𝐢𝐭𝐡 𝐭𝐡𝐞 𝐆𝐥𝐨𝐛𝐚𝐥 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞—𝐰𝐡𝐞𝐫𝐞 𝐯𝐢𝐬𝐢𝐨𝐧 𝐦𝐞𝐞𝐭𝐬 𝐯𝐚𝐥𝐮𝐞, 𝐚𝐧𝐝 𝐞𝐯𝐞𝐫𝐲 𝐫𝐨𝐥𝐞 𝐢𝐬 𝐞𝐦𝐩𝐨𝐰𝐞𝐫𝐞𝐝 𝐟𝐨𝐫 𝐬𝐮𝐜𝐜𝐞𝐬𝐬!

09/30/2025

AI’s Double-Edged Sword

In this featured video, Dr. Fei-Fei Li—renowned as the “Godmother of AI”—shares her deeply influential vision for the future of artificial intelligence. As a professor of computer science at Stanford University and co-director of Stanford’s Human-Centered AI Institute, Fei-Fei Li brings a uniquely balanced perspective: she champions AI as a transformative civilizational tool, but urges caution against both hype and fear.

Fei-Fei Li’s central message is that AI must remain a double-edged sword—powerful, but requiring careful human guidance. She sees the greatest promise in using AI to enhance human creativity, revolutionize education, empower scientists and artists, and tackle repetitive or dangerous tasks. Dr. Li emphasizes that as AI grows, society must avoid simply projecting human intelligence onto mathematical systems, and should focus on building technology that truly augments what people do best.

Through this lens, Li is a passionate advocate for reimagining how we teach and learn. She argues that education systems should go beyond rote memorization and instead leverage AI to nurture creativity, adaptability, and curiosity in future generations. Her call is for thoughtful, human-centered AI design—making sure these tools serve and empower people, operate ethically, and respect human values.

Fei-Fei Li also stresses the importance of interdisciplinary collaboration, inclusive governance, and social responsibility; she challenges technologists and policymakers alike to ensure that AI increases opportunity and wellbeing for all. This interview provides both inspiration and practical wisdom—urging us to harness AI for good, while always keeping humanity at the center.

09/25/2025

𝐎𝐯𝐞𝐫𝐜𝐨𝐦𝐢𝐧𝐠 𝐂𝐨𝐠𝐧𝐢𝐭𝐢𝐯𝐞 𝐓𝐫𝐚𝐩𝐬 𝐢𝐧 𝐀𝐈 𝐏𝐫𝐨𝐣𝐞𝐜𝐭𝐬—𝐆𝐮𝐢𝐝𝐚𝐧𝐜𝐞 𝐟𝐫𝐨𝐦 𝐭𝐡𝐞 𝐆𝐥𝐨𝐛𝐚𝐥 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞

Even with cutting-edge technology, far too many AI projects fail—not because of software or algorithms, but due to persistent cognitive traps in how projects are planned and evaluated. At the Global Institute of Data Science (GIDS), we educate and empower organizations to spot these traps and transform AI outcomes.

⚡ 𝐁𝐞𝐰𝐚𝐫𝐞 𝐭𝐡𝐞 𝐁𝐢𝐠 𝐓𝐡𝐫𝐞𝐞 𝐂𝐨𝐠𝐧𝐢𝐭𝐢𝐯𝐞 𝐓𝐫𝐚𝐩𝐬

⚡ 𝐀𝐧𝐭𝐡𝐫𝐨𝐩𝐨𝐦𝐨𝐫𝐩𝐡𝐢𝐬𝐦
Don’t treat algorithms like people. Projecting human intelligence onto mathematical models sets unrealistic expectations—and leads to poor project planning.

💡 𝐏𝐫𝐞𝐜𝐢𝐬𝐢𝐨𝐧 𝐈𝐥𝐥𝐮𝐬𝐢𝐨𝐧
Don’t mistake mathematical confidence for true accuracy. Just because your model returns highly confident numbers doesn’t mean those results are always reliable. Like a weather app, AI forecasts need careful human interpretation.

📚 𝐎𝐯𝐞𝐫𝐜𝐨𝐧𝐟𝐢𝐝𝐞𝐧𝐜𝐞 𝐁𝐢𝐚𝐬
Confidence without oversight causes 67% of the 85% failure rate in AI projects. Always double-check and question results. Build in processes for broad review, scenario testing, and real-world validation.

🤖 𝐁𝐞𝐬𝐭 𝐏𝐫𝐚𝐜𝐭𝐢𝐜𝐞𝐬 𝐟𝐨𝐫 𝐄𝐧𝐝 𝐔𝐬𝐞𝐫𝐬

🤖 𝐓𝐫𝐞𝐚𝐭 𝐀𝐈 𝐚𝐬 𝐚 𝐓𝐨𝐨𝐥, 𝐍𝐨𝐭 𝐚 𝐂𝐨𝐥𝐥𝐞𝐚𝐠𝐮𝐞:
Apply the same rigor and skepticism as you would to any enterprise software or analytics solution.

🚦 𝐑𝐞𝐪𝐮𝐢𝐫𝐞 𝐎𝐯𝐞𝐫𝐬𝐢𝐠𝐡𝐭 𝐚𝐧𝐝 𝐈𝐧𝐭𝐞𝐫𝐩𝐫𝐞𝐭𝐚𝐭𝐢𝐨𝐧:
Use human judgment to review outputs, set boundaries, and catch subtle errors algorithms may miss.

🔍 𝐄𝐝𝐮𝐜𝐚𝐭𝐞 𝐓𝐞𝐚𝐦𝐬 𝐨𝐧 𝐁𝐢𝐚𝐬𝐞𝐬:
Raise awareness of these common traps across departments—finance, operations, IT, and leadership.

🎓 𝐓𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦 𝐰𝐢𝐭𝐡 𝐭𝐡𝐞 𝐆𝐥𝐨𝐛𝐚𝐥 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞
GIDS brings expert evaluation, training, and governance frameworks to help organizations unlock AI’s value responsibly, avoid cognitive traps, and achieve meaningful project outcomes.

𝐑𝐞𝐚𝐝𝐲 𝐭𝐨 𝐭𝐮𝐫𝐧 𝐢𝐧𝐬𝐢𝐠𝐡𝐭 𝐢𝐧𝐭𝐨 𝐢𝐦𝐩𝐚𝐜𝐭? 𝐂𝐨𝐥𝐥𝐚𝐛𝐨𝐫𝐚𝐭𝐞 𝐰𝐢𝐭𝐡 𝐭𝐡𝐞 𝐆𝐥𝐨𝐛𝐚𝐥 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞 𝐚𝐧𝐝 𝐦𝐨𝐯𝐞 𝐲𝐨𝐮𝐫 𝐀𝐈 𝐩𝐫𝐨𝐣𝐞𝐜𝐭𝐬 𝐟𝐫𝐨𝐦 𝐫𝐢𝐬𝐤 𝐭𝐨 𝐫𝐞𝐬𝐮𝐥𝐭𝐬.