AI reads 2,000 historical illustrations: individual errors, near-perfect aggregate patterns
_akpiper · x · 2026-11-23
An EMNLP 2026 paper tests GPT-4.1, Gemini 2.5 Flash, Claude Sonnet 4.5, Qwen3-VL-8B and LLaVA-13B on 2,000 historical children's book illustrations against 700k+ crowd annotations. Key findings: 97–98% accuracy on illustration detection; models systematically underestimate emotional intensity; and crucially, even with individual errors, models reproduce corpus-level feature distributions with correlations up to 0.99 — making them viable tools for large-scale visual culture research.
More from Multimodal
- w2v-bert-fast open-sources exact 3x speedup for w2v-BERT 2.0 training and inference — kastnerkyle · 2026-11-23
- Free Claude plugin turns your MCP-connected video models into an MV factory — AIandDesign · 2026-11-23
- AI short film 'The Wow Lunch': Midjourney images animated by Seedance 2.5 via Claude MCP — azed_ai · 2026-11-22
- Wan 2.2 Image-to-Video on 8GB VRAM: 4-Step Clips With Auto Last-Frame Saving — Equivalent-Ad1959 · 2026-11-22
- Zero Custom Nodes: A ComfyUI Pipeline That Puts a Model Into Any Reference Video's Motion — lmcdesign · 2026-11-22
- Open-source Short-Drama Skill helped a creator ship 20+ AI dramas with 50M+ views — vista8 · 2026-11-22