It is with great pride that we are today highlighting the exceptional work of three brilliant students from Ensae Paris: Raphaël Dabadie, Alexandre Combes and Mehdi Sfar.

At just 23 years of age, these undergraduate talents have created a revolutionary new benchmark for assessing the cognitive abilities of artificial intelligences, more specifically their social and behavioral intelligence. Their project, dubbed "Probing LLM social intelligence via Werewolf" (see https://were wolf.foaster.ai/), has not only captivated the global scientific community, it has also rattled American tech giants such as OpenAI and Grok!

Imagine: a research paper co-written by these young prodigies, developed within their startup Foaster.ai, analyzing the psychology of AI models through the game of Wolf-Werewolf. A totally new approach! By simulating hundreds of games in which AIs embody villagers, werewolves or other roles, they have dissected the personalities of the models: charismatic leaders who unite players, manipulative bluffers ready to betray their allies, or paranoids leading to their own downfall. This benchmark not only ranks AIs in terms of social intelligence, strategy and manipulation, it also provides laboratories with a valuable tool for refining future models by integrating more realistic social dynamics.

And the results? Spectacular! The paper was shared on X (formerly Twitter) by none other than OpenAI's president and co-founder himself. Over 300,000 views in the blink of an eye, hundreds of likes and shares from top OpenAI researchers, Google DeepMind, Mistral AI and NVIDIA. The icing on the cake: xAI's co-founder, Elon Musk's right-hand man, contacted them personally to ask for Grok 4 to be added to the ranking!

This recognition has propelled Foaster.ai onto the international AI scene, reinforcing its legitimacy and visibility.

Who could have predicted that a handful of French students, armed with their expertise in data science and statistics acquired at ENSAE, would attract the interest of the world's leading laboratories? This success is no accident. It is a perfect illustration of the excellence of the training provided at ENSAE Paris, where we cultivate not only cutting-edge technical skills in artificial intelligence, machine learning and data analysis, but also the creativity and audacity to push back the frontiers of tech.

Raphaël, Alexandre and Mehdi embody this new generation of ensaeologists who are not content to just learn: they innovate and impact the real world. Their work, the fruit of months of development and meticulous analysis, opens up fascinating perspectives for the future of AI. By popularizing the personalities of the models, they are not only helping researchers to create more ethical and social AIs, they are also positioning France as a key player in this burgeoning field.

For more information: https: //werewolf.foaster.ai/