What 81,000 people want from AI!

The rapid advancements in artificial intelligence, particularly in large language models (LLMs), have brought to the forefront a critical challenge: aligning AI system behavior with human values, preferences, and intentions. While significant progress has been made in optimizing models for performance metrics such such as perplexity or accuracy on benchmark datasets, the ultimate utility and safety of these systems hinges on their ability to interact in a manner that is perceived as helpful, harmless, and honest by human users. The empirical insights derived from large-scale user studies are therefore invaluable, shifting the discourse from theoretical alignment principles to actionable engineering requirements. Anthropic's initiative to conduct 81,000 interviews regarding user preferences for AI systems represents a substantial effort to gather this critical empirical data. This study transcends the anecdotal and moves towards a statistically significant understanding of what a divers