Anthropic found out how language changes AI responses.

anthropic.com/research/claude-values-models-languages Imagine two people presenting the same business plan to a neural network. One writes in Hindi and is likely to receive encouraging feedback praising its strengths. The other writes in Russian and is more likely to see an analysis of its weaknesses and questions about the numbers. The request is identical, the model is the same, but the plan's evaluation may be different. This isn't a hypothetical scenario, but an example from research: the co