Deep dives into human-centered AI development, preference modeling, and the future of human-AI collaboration
Through my work in data annotation, I discovered that when AI experts disagree on the "correct" answer, we're not seeing a data quality problem—we're seeing human values in action. This insight fundamentally changes how we should approach AI development.
During a creative writing AI evaluation project, five PhD-level annotators provided completely different "correct" responses to the same prompt. Initially, this looked like a problem to solve. But it wasn't—it was the most important signal in the entire dataset.
How to capture nuanced human preferences that vary by context, culture, and individual needs. A framework for moving beyond binary "good/bad" metrics.
Why traditional AI metrics miss what actually matters to users, and how product managers can build better success frameworks for human-centered AI.
A deep dive into how expert disagreement in AI training data reveals crucial insights about human values and preference diversity.
Most organizations ask "What AI tools should we use?" when they should ask "Are we ready for AI to succeed?" A framework for building sustainable AI capabilities.
Applying traditional PM principles to AI projects, with lessons learned from managing cross-functional teams building human-centered AI products.
Why Cursor + Claude Code represents a paradigm shift in software development, and what it means for context engineering, rapid prototyping, and architectural thinking.