Nonprofits

Independent research institute founded in 2017 producing policy research on the social impact of AI and industry power.

Fully open models with their data and training code: OLMo, Tulu, Molmo, and the Dolma corpus.

Independent non-profit AI institute in Edmonton, begun in 2002 as AICML and one of three Pan-Canadian AI Strategy institutes.

Berkeley non-profit founded by Paul Christiano in 2021, working on theory for mechanistic explanations of neural networks.

AI safety organization focused on evaluating and monitoring scheming and deception in frontier AI systems.

The BGE embedding models used across open retrieval stacks, the Emu multimodal models and FlagOpen.

San Francisco nonprofit doing technical AI safety research and field-building; organized the 2023 Statement on AI Risk.

AI governance think tank founded at Yale in 2016, now an independent non-profit with offices in London and Washington, DC.

501(c)(3) nonprofit that crawls the web monthly and publishes a free open archive of web data collected since 2008.

Independent, community-rooted AI research institute founded by Timnit Gebru in December 2021.

Grassroots open research: The Pile, GPT-NeoX and Pythia, and the LM Evaluation Harness everyone uses.

Data on AI's trajectory: training compute trends, data limits, and the FrontierMath benchmark.

Pan-European AI network of excellence founded in 2018, with institutes and units at research institutions across Europe.

AI security research and adoption non-profit, known for adversarial attacks on superhuman Go AIs and the tuned lens.

German AI research center founded in 1988 as a non-profit public-private partnership, with sites across Germany.

Open-science lab funded by Xavier Niel and partners: Moshi real-time voice and Hibiki live translation.

Open image-text datasets (LAION-400M, LAION-5B) that trained Stable Diffusion and OpenCLIP.

Berkeley nonprofit founded in 2000 as the Singularity Institute, focused on existential risk from superintelligent AI.

Evaluating dangerous capabilities: the time-horizon measure of how long a task AI agents can finish.

Nonprofit AI safety and security research organization that pioneered the AI control agenda.

Non-profit archive of publicly available source code, launched by Inria in 2016 with UNESCO support.

UK national institute for data science and AI, an independent charity based at the British Library, founded in 2015.

Independent nonprofit research lab building open tools for public oversight and understanding of AI systems.

Toronto's independent AI institute, home to deep learning researchers and work such as neural ODEs.

Independent non-profit founded in 2018 building AI for underserved communities in health, agriculture and education.