Geo founder Yaniv Tal has identified four weaknesses in online information that make AI answers unreliable: lost provenance, flattened authority, hidden disagreement, and repeated model-generated errors. According to Tal, these weaknesses arise from the internet's design, which prioritizes information distribution over preserving authority, origin, or accountability.
Tal argues that 'AI doesn’t have a truth problem, the internet does,' and that information loses critical context as it is scraped and republished across websites. This can lead to models treating material with different standards of evidence as if it carries similar weight.
Addressing Weaknesses
Geo's proposed solution involves separating a statement from its author and the evidence supporting it, and ordering arguments by their assessed strength while retaining competing material. The platform organizes information through independent communities called Spaces, where members can contribute and participate in discussions, and people with relevant knowledge can apply to become editors.
Human Verification
Tal believes that human judgment is the scarce input in verifying AI answers, and that machines generate more content than people can process. To address this, Geo relies on contribution records and domain-specific communities, rather than allowing anonymous material to carry the same status as claims attached to people with public histories. Editors apply through individual Spaces, and members take part in governing the subjects they follow.
Expertise and Accountability
Tal says expertise will not be assigned by one central authority under the proposed system, but rather contributors will build a reputation through their work. Geo intends for this record to follow a person between Spaces, providing a level of accountability and transparency. By leveraging human verification, Geo aims to create a more reliable and trustworthy knowledge network.
