Compar:IA

[Compar:IA](https://comparia.beta.gouv.fr/) is a conversational AI comparator (a chatbot arena) developed within the Ministry of Culture with a dual mission: * Train and raise awareness of model pluralism, cultural and linguistic biases, and environmental issues of conversational AI. * Build and publish datasets of questions and preferences useful for observing usages and improving conversational AI in French. The comparator:IA is developed as part of the compar:IA state start-up (incubator of the [Digital Workshop](https://www.culture.gouv.fr/Thematiques/innovation-numérique/Aides-a-l-innovation-et-a-la-transformation-digérique/L-Atelier-digérique #:~:text=L'Atelier%20num%C3%A9rique%20est%20l,engage%20personally%20for%20le%20r%C3%A9soudre.) and [AllIAnce](https://alliance.numérique.gouv.fr/) integrated into the [beta.gouv.fr](https://beta.gouv.fr) programme of the [Interministerial Directorate of Digital (DINUM)](https://www.numérique.gouv.fr/dinum/) which helps public administrations to build useful and easy-to-use digital services. Link to the arena: comparia.beta.gouv.fr Link to the Github repository of the project: <https://github.com/betagouv/ComparIA> For more information: <https://www.comparia.beta.gouv.fr/a-propos> The data collected via compar:IA are published in three datasets: * **Compar conversations:IA**: Dataset of all questions asked and answers received on the compar:IA platform. * **Votes compar:IA**: A dataset of preferences expressed by users after an entire conversation. Voting makes it possible to compare two models when no reaction has been given on individual messages. * **Compar:IA******:** Dataset of compar:IA user reactions at the message scale. It reflects the preferences expressed during the discussions, message by message.