Coverage by Theo Lindgren for Kernel. The overview below summarizes the reporting; read the original for the full analysis.
What the coverage examines
Kernel reports on an evaluation of religious and ethical bias using AllFaith, developed by a consortium drawing on Baylor, Brigham Young, Notre Dame, and Yeshiva universities. Aligned AI’s evaluation recorded 9% of the maximum possible conversion bias across 14 faiths and included religious perspectives in 92% of relevant responses.
How to read the results
The article also explores Aligned’s approach to coordinating multiple models. It notes that Aligned’s evaluation and the consortium’s leaderboard used separate runs: scores can vary with the run and judging model. These findings describe performance on a particular evaluation, not a guarantee of neutrality or accuracy in every answer.
Read the original article on Kernel ↗
About Aligned AI
Aligned AI builds sovereign AI using open-source models and infrastructure it operates. Its products bring AI assistance to individuals and organizations, with an emphasis on privacy, control, and accessibility.
Media inquiries: press@joinaligned.ai