Sleeping AI began as an experiment inspired by Nicolas Bourbaki: a collective identity for talented researchers, engineers, and hackers with a fierce commitment to releasing state-of-the-art infrastructure and research. We want to build projects people use—not work that ends as one more citation or experiment.

We began as a small group spanning Germany, the United States, Montenegro, and England. As the collective grew, contributors joined us from Russia, Israel, Belgium, and China. Sustaining international collaboration through geopolitical strain has not always been easy, but it remains essential to who we are.

Our ethical position is simple: external pressure or internal resistance should not determine who may contribute or collaborate, provided the work serves our research direction and respects the people involved. We value technical ability, thoughtful engineering, and curiosity.

To be clear, we are not trying to become DeepMind, LAION, OpenAI, or a startup built to raise millions. We do not have the resources or facilities of large laboratories—and we do not want the institutional weight that can make a group slow and inflexible. We choose the fundamental constraint of staying organised and small.

We are individuals from different backgrounds, cultures, ethical traditions, and narratives, brought together by a desire to contribute research and help shape humanity’s future. None of us will remain biological entities forever, but our contributions can outlive us.

We identify with the spirit of the early internet and with moments such as the 2012 internet blackout associated with Aaron Swartz and many others. Artificial intelligence is still young. We want to help protect what is fundamental while its norms are still being formed.

Information belongs to everyone. Don’t let unjust rules dictate what is fair.

We do not have a polished mission statement. Funding us will not turn anyone into a Wall Street bull; we do not generate revenue. We write papers because a world shaped by lawyers, corrupt journalism, unjust laws, and segregation by belief still needs people willing to diverge—and to contribute something enduring to humanity.

Yes, you are welcome to criticise us.

Thankfully,
Sleeping AI

Datasets are pivotal to progress in artificial intelligence, especially high-quality and aesthetically relevant collections comparable with those released by LAION and others. We have developed and released datasets in areas where the market and research field were not yet mature, sometimes before public concerns were fully understood.

We are interested in collecting the world, but not in scale for its own sake. We focus on quality and on datasets capable of meaningfully advancing artificial intelligence. Some of our releases have drawn public criticism. We have not always communicated them well enough, but our intent has been consistent: to make open source what large laboratories often do behind closed doors.

Our aim is to make the field flatter and give more people the opportunity to develop state-of-the-art models. A core principle is compliance with the laws that apply to our work. As a not-for-profit laboratory, we also have the freedom to organise datasets that a commercial entity may choose not to pursue.

Since August 2026, every Sleeping AI dataset paper must include a data-release statement. We welcome questions and criticism, and encourage people to focus that scrutiny on specific projects rather than targeting the laboratory or its contributors.

We are also creating public channels for accountability: social networks for responding to concerns, open AMAs, a Discord server, and forums where people can discuss our projects and make their voices heard.