An AI safety team is designing a content moderation system for a major social platform. Claude must classify user posts into: Safe, Borderline, Harmful, Severely-Harmful. The 'Borderline' category is the primary accuracy challenge — posts that are provocative but legal, potentially triggering but not policy-violating. Currently, Claude classifies 8% of posts as Borderline. Human reviewers classify 3% as Borderline, and 5% of Claude's Borderline classifications are later escalated to Harmful. The team wants to calibrate Claude's Borderline classification to match human reviewer standards. Which is the most rigorous calibration approach?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — systematic calibration study with confusion matrix analysis is the most rigorous approach for aligning Claude's Borderline classification with human reviewer standards. The current data (8% vs 3% Borderline, 5% Borderline→Harmful escalation rate) provides initial signals but not enough to identify specific error modes.
Full explanation below image
Full Explanation
Systematic calibration study with confusion matrix analysis is the most rigorous approach for aligning Claude's Borderline classification with human reviewer standards. The current data (8% vs 3% Borderline, 5% Borderline→Harmful escalation rate) provides initial signals but not enough to identify specific error modes. The calibration study: (1) sample 500 posts from each Claude-assigned category, (2) have human reviewers classify independently, (3) compute the full confusion matrix — what does Claude call Borderline that humans call Safe? What does Claude call Safe that humans call Harmful?, (4) analyze the specific content types driving each disagreement cluster, (5) generate few-shot examples that directly address the identified error patterns. This approach is rigorous because it identifies the specific content patterns where Claude diverges from human judgment, enabling targeted correction rather than general instruction. Option A (add examples for Claude's False Positives) addresses one error direction but ignores the 5% of Claude's Borderline being Harmful (False Negatives). Option B (confidence routing) doesn't calibrate Claude's classification — it routes uncertain cases to humans, which is a valid operational strategy but not a calibration approach. Option D (continuous harm score) is an interesting alternative architecture but changing from classification to regression is a major architectural change, not a calibration approach.