Wiki Is Not Enough [Information Architecture for LLMs] (replicated.live)

Google has introduced the Open Knowledge Format (OKF), an innovative standard for structuring knowledge in a way that enhances the effectiveness of large language models (LLMs). By utilizing a simple wiki stored in a git repository, which incorporates basic directory naming and index files alongside minimal key-value metadata, Google aims to simplify the integration of information. This approach allows for a more organized flow of relevant data to keep LLMs focused within a manageable context window, although the limitations of context length—ranging from 50K to 250K tokens—pose challenges for larger, ongoing projects. The significance of OKF for the AI/ML community lies in its potential to mitigate common issues associated with LLMs, such as the accumulation of stale or contradictory information in knowledge bases. As projects evolve, the ability to absorb and implement feedback seamlessly is crucial for maintaining quality and coherence. The proposed architecture effectively allows for dynamic handling of increasing data volumes, with advanced features like permalinks to maintain context and coherence across multiple tiers of information. This structured approach not only streamlines knowledge management but could also help teams avoid the pitfalls of "information rot" that hamper the effectiveness of LLM-generated wikis.