DatBench fixes VLM evals: 70% blindly solvable, 42% mislabeled, 35% prod gap Posted by hurrycane on Tuesday January 06, 2026 @12:54PM from the datologyai.com dept. https://www.datologyai.com/blog/datbench-discriminative-faithful-and-efficient-vision-language-model-evaluations?source=HN ( no comments | 5 points | on HN )
There are no comments yet.