# User-owned Foundation Models

- Channel: [ZuBerlin](https://streameth.org/zuberlin)
- Date: 2024-06-18
- Duration: 25:14
- Topics: decentralized AI, foundation models, data ownership, user data, AI governance
- Watch: https://streameth.org/watch/6671866d07f92b086c3b62a6
- Download: https://vod-cdn.lp-playback.studio/raw/jxf4iblf6wlsyor6526t4tcmtmqa/catalyst-vod-com/hls/7b7533k51m0hpioh/1080p0.mp4

## Description

Clip The video focuses on the development of user-owned foundation models in AI, particularly in the context of the speaker's work at Vana. It begins with a discussion on why controlling AI is powerful, citing examples of AI-generated content that could shape perceptions of truth. The talk then delves into the creation of AI models using public data and highlights the challenges faced due to limited publicly available data.

The speaker discusses how users legally own their data and proposes that if a large number of users pooled their data, it could be used to train competitive AI models. They provide calculations to show that user-generated data could surpass the amount used in current frontier models and suggest that decentralized compute resources, like those from cryptocurrency miners, could be mobilized for training these large-scale models.

The video also touches on financial incentives for contributing high-quality data and covers various methods for ensuring data quality and authenticity. It explores governance issues around user-owned models, including how contributors might be compensated and how usage rules could be set collectively.

Finally, the speaker encourages actions such as building data liquidity pools, integrating advanced privacy frameworks, running validators to check data quality, and supporting federated machine learning approaches. The overarching theme is empowering users to contribute to and control AI models as an alternative to relying on big tech companies' leftovers