Blog
ALL POSTS
Essays and articles on AI, data, and the infrastructure underneath.
AI: From the Engine Room
A DATA LEADER'S PERSPECTIVE ON MODERN AI
AI Mechanisms
I've Heard This Engine Before
I worked on attention mechanisms before ChatGPT made them famous. That perspective has shaped how I separate AI signal from noise.
The Attention Mechanism Explained
"Attention" is a brilliant name for what these models do, and also misleading in ways that matter.
Training Economics
"Let's just train a model on our data" sounds straightforward. The actual cost structure tells a different story.
The Efficiency Illusion
That benchmark showing the smaller model matches GPT-4? Probably accurate. Also hiding tradeoffs that only surface in production.
Context Windows and Memory
The most common misconception I encounter: people assume these models remember previous conversations. They don't.
The Proprietary Data Paradox
The Multiplier Effect
Proprietary data isn't the AI advantage most organizations think it is. What that data connects to is.
The Tacit Knowledge Bottleneck
The most valuable knowledge in any organization is often the hardest to capture, because the people who have it can't fully explain it.
Why Vectors Aren't Enough
Vector search finds what's similar. Expertise often requires connecting things that aren't similar at all.
UI as the Ultimate Guardrail
AI systems speak with uniform confidence whether they're right or wrong. Interface design determines whether that difference becomes visible.
Forward-Looking Governance
Hallucination is a Feature, Not a Bug
Vendors frame hallucination as a bug they're fixing. It's actually inherent to how these systems work, which changes the governance question entirely.
Working With the Machine
The most valuable prompting skill isn't writing better prompts. It's knowing when prompting isn't the answer.
AI Readiness Is a Governance Question
Every AI conversation eventually becomes a data conversation. Every data conversation eventually becomes a governance conversation. The sequence matters.
What the Engine Room Taught Me
These systems are more comprehensible than marketing suggests, and more limited than hype implies. Both are true simultaneously.