Measuring Social Biases in Grounded Vision and Language Embeddings — mentioned by 0xab

Wed, 04 Jun 2025 01:00:28 GMT

> When VLMs make errors, they don't make random mistakes. Instead, 75.70% of all errors are "bias-aligned" - meaning they give the expected answer based on prior knowledge rather than what they actually see in the image. Yeah, that's exactly what our paper said 5 years ago! They didn't even cite us :( "Measuring Social Biases in Grounded Vision and Language Embeddings" https://arxiv.org/pdf/2002.08911

Paper and discussion history. A mention is not an endorsement.

Measuring Social Biases in Grounded Vision and Language Embeddings · HackerPapers discussions

Measuring Social Biases in Grounded Vision and Language Embeddings — mentioned by 0xab