AI Design Evaluation: TASTE vs ACM 3698105
Comparative Analysis: TASTE vs. ACM 3698105 Overview Two frameworks for evaluating AI-generated visual content — compared across purpose, domain, dimensions, signal quality, and practical use. TASTE (arXiv 2605.20731) ACM 3698105 Published May 2026 ACM Computing Surveys, Nov 2024 Domain Commercial graphic design (posters, ads, social media, logos) AI-generated fine art (painting, illustration, style transfer) Purpose Preference dataset + benchmark for training/evaluating reward models Unified evaluation rubric + taxonomy survey Annotation 10 professional designers, pairwise rankings, 1,600 ratings/criterion User study validating rubric weightings Prompt-based use Usable as rubric but <0.55 agreement without fine-tuning Works directly as a scoring checklist — no training needed Evaluation Dimensions TASTE — 9 criteria across 2 cohorts Aesthetics Cohort (evaluates on aesthetic feel): ...