Interactive · Week 3 · Section 1

Should you actually fine-tune?

Most teams that fine-tune in 2026 didn't need to, and most teams that needed to didn't realize it. Describe your task profile below — task type, quality bar, volume, latency, dataset size, domain, and eval maturity — and the matrix will rank the seven main approaches with the actual reasoning behind each score. The default profile is the one most teams reach for fine-tuning on; watch what actually wins.

Task profile

What's happening here

Scoring uses transparent, additive rules — every + or − shown on a card corresponds to a single explainable rule. There is no ML or fitting. The intent is pedagogical: see why each approach scores the way it does, and develop intuition for which task profiles push toward which solutions. Real decisions involve more dimensions (regulatory, team capability, vendor lock-in, etc.); this widget covers the seven that matter most. Source-of-truth lives as <FineTuneDecisionMatrix/> in the curriculum codebase.