AI article
Can an LLM measure UI thresholds from screenshots? 164 crops vs DOM gold labels
Community description: 12 pages × 10 UI rules measured in headless Chrome as gold, then judged by GPT and Gemini from 1:1 crops only. On the 60 pairs with real contrast: GPT 54/60, Gemini 41/60. All six GPT misses were 15px vs 13.5px; the 44px rules could not test threshold discrimination — and my first tally hid that.
Dev.to | Sep 21, 2026 | Sumitsuke
Automated excerpt
GPT: pasted by hand into the ChatGPT UI, one page per submission, 12–15 images at once. All six GPT misses are R1: three pages at 13. 5px answered NO, three pages at 15px answered YES. H1 "visual rules are caught by both judges" — GPT yes (36/36), Gemini no (FN 44%).
Selected automatically from source text; not independently written or fact-checked. Read the original for full context.