9 minAI Deep Dives
Factor 1000: Why a Vision Model Misreads Receipts
A bakeoff across 16 real receipts exposed a systematic factor-1000 error. Why a deterministic guard sits above the model, not inside it.
5 posts on this topic.
A bakeoff across 16 real receipts exposed a systematic factor-1000 error. Why a deterministic guard sits above the model, not inside it.
Established companies with approval processes lag on AI not from inertia, but due to a growing gap between what tools can do and what their structures allow.
AI-assisted development has shifted its bottleneck from writing code to judging it. Why verification now matters, from daily practice, not theory.
Three practical conclusions from the 2026 Gartner and HBR work-trend reports, viewed through my own work with AI agents.
A practical starting point for small and medium enterprises: choose one process, inspect the data, define a test, and keep a human decision point.