Cohere AI Updates: August 21, 2026

1. Cohere Labs Finds Cultural Diversity Is Lost in Post-Training

Cohere. Cohere Labs published research arguing that cultural diversity is frequently lost during the post-training data mixes used for modern large language models. The analysis finds explicit cultural signals decline sharply during post-training while geographically concentrated, task-specialized data dominates. The team released a culturally tagged dataset alongside the writeup for further study. Source