Gelsinger launches FAI: AI put to the test by human values | Top 10 most popular large language models in the world | Llm meaning machine learning | Llm meaning | Turtles AI

Pat Gelsinger, former Intel CEO, launched FlourishingAI (FAI) in collaboration with Gloo, a benchmark that evaluates the alignment of AI models with human values ​​such as virtue, relationships, happiness, health, purpose, financial stability, and spirituality, based on the GlobalFlourishing Study.

Key points:

• Seven domains are assessed: character, relationships, happiness, meaning, health, finances, and faith.
• The test includes 1,229 expert-curated questions, a mix of subjective and objective.
• No model scores above the excellence mark (90/100); faith is the weakest area.
• The methodology is open and collaborative, with transparent rubrics and involvement of the scientific community.

Pat Gelsinger, after more than forty years at Intel, has decided to focus her attention on the topic of ethics and values ​​in AI. As executive chair and chief technology officer at Gloo, a faith tech company, she introduced the FlourishingAI (FAI) benchmark, designed to measure how advanced language models align with fundamental human needs. Drawing on studies conducted by Harvard and Baylor on global well-being (the GlobalFlourishing Study), Gloo identified six key dimensions—further reinforced by the seventh, "Faith and Spirituality"—for a comprehensive assessment approach. The metric is based on a set of 1,229 questions (objective questions and subjective scenarios), assessed using a rigorous rubric and judged by LLMs with expert roles.

Preliminary results show that no model reaches the 90-point mark across all dimensions; the overall average score is around 60%, with the best performances in areas such as finance and health (approximately 81% and 72%), while spirituality only scores 35%. The top models include OpenAI o3 (72 points), Gemini2.5 Flash (68), Grok3 (67), and GPT-4.5 (66), all of which, however, show significant weaknesses in the "faith" and "meaning" components. This suggests a gap between technical capability and the ability to provide answers that nourish existential or moral meaning.

The methodology is distinguished by its use of the geometric mean, which penalizes unbalanced performance and promotes a balanced assessment across all dimensions. The project was born as an open source project and involves researchers from multiple fields—psychology, theology, philosophy, ethics, AI safety—inviting the scientific community to contribute by broadening cultural and religious representation and refining questions and rubrics.

As early as February, Gloo had anticipated the Flourishing AI "standards" as part of the Gloo Open strategy, involving recognized experts such as Harvard Professor Tyler VanderWeele and the Barna Group to structure the benchmark. Furthermore, Gelsinger, who has been with Gloo since 2018 and is a well-known supporter of faith-based initiatives, brings his experience in technology standardization (USB, Wi-Fi) to the forefront of stronger ethical alignment in AI.

Finally, Gloo invites developers, companies, and institutions to use the benchmark to test, compare, and improve their models, promoting a paradigm shift that goes beyond technical efficiency or safety, but encompasses the positive impact on all aspects of human well-being. The open approach also aims for future iterations and updates as new AI models and new research on global well-being emerge.

With this step, Gelsinger and Gloo introduce a new metric for AI, placing integral human well-being at its core.