Developing large language technologies as a public service tool can revolutionize the way individuals interact with information, improving access to knowledge and resources. Public AI is a movement and vision for open-source AI that emphasizes sustainability, optimizes for social impact and reduces reliance on monopolistic and proprietary models.

By prioritizing sovereignty and neutrality, we can try to ensure the service remains unbiased, let users understand transparently how their data is handled, how governance decisions are made. Ultimately, this gives individuals more agency over how they use and build on AI. To tackle this complex challenge, research and development can begin with a variety of datasets and resources. For instance, using large, publicly available datasets like Common Crawl, FineWeb 2, Wikipedia, news archives or legal texts, to provide a comprehensive foundation. Additionally, there are APIs like OpenRouter or Hugging Face that offer pre-trained models under permissive licenses—these can be used in conjunction with a pooled community effort to tune and customize a model. It's also essential to incorporate multilingual data to reflect global accessibility and inclusivity, leveraging resources from platforms like OpenSubtitles for translation and various global news services for content variety.

Implementing a large language model as a public service tool could be transformative, impacting society in several areas. By focusing on sustainability and transparency to minimize environmental and political biases, the tool can become a trusted source of information. It could democratize access to knowledge, bridging gaps for individuals in underserved areas or with lesser economic resources, offering clear explanations or translations at a personal level. Moreover, ensuring user agency means that individuals can interact with the AI in a safe, informed manner—understanding the potential risks of AI while benefiting from its capabilities. This approach could shift the landscape for how AI is developed, used, and governed in the public sphere, leading to more inclusive, responsible, and community-driven technological advancements.

🅰️ℹ️ generated with APERTUS-70B-INSTRUCT

HACKnight Goals

🟢 beginner

Try the public inference at chat.publicai.co. Create an account, and test the Apertus ("Switzerland") instance for yourself. Please take five minutes to fill out the Apertus User Satisfaction Survey with your feedback.

🔵 intermediate

Learn more about the mission of Public AI, and how we plan to further support the Inference Utility based on the Apertus project. Brainstorm and share ideas of cool projects that could be made possible with this. Join our new Hugging Face organization, and help to build a local community. There was a presentation in Biel/Bienne yesterday (in French).

⬤ advanced

Find more technical background and links about Apertus on the Hackathon resources. Connect to the Public AI API with your own app (an example is here), or even run the model on your own server/hardware (see blog post). Learn to connect your model to an MCP service, or even write your own for the MCP 1st birthday hackathon.