"This small survey demonstrates that leveraging synthetic data and running evals is moving from niche to mainstream."

A survey of 413 AI practitioners reveals extensive experimentation with AI techniques, highlighting the dominance of open-source models and internal AI applications. Key findings include widespread evaluations, significant performance gains from RLFT, and mainstream fine-tuning practices, particularly among larger firms. Synthetic data is commonly used for evaluations, and operational focuses include interaction management and context improvement. The survey also details demographic data, company sizes, and sector distributions among respondents.