Knowledge Representation Insights
Kelvin shares insights on the challenges of knowledge representation, highlighting the trade-off between centralization and coverage. He emphasizes the importance of dense models for managing and controlling representations in various applications.In this clip
From this podcast
No Priors
No Priors Ep. 15 | With Kelvin Guu, Staff Research Scientist, Google Brain
Related Questions
Tell me about knowledge representation
I'm thinking that classical or deep ML solutions have a flaw in that they cannot be extended. For example, if one builds a model and wants to introduce a new feature, typically the model has to be retrained from scratch. So, I'm considering building a Knowledge Graph using LLMs. This knowledge graph would have to include time-dependent data (for example, it should be able to retrieve the "current" President of the USA and also previous presidents if asked). I'm thinking this Knowledge Graph could be used for Retrieval Augmented Generation (RAG) to help with business goals or maybe used with more clever User Interfaces (UIs). I'm not sure how to build or populate this KG and also have a rough idea of how to use it. Can you help me with suggesting particular paths for building and populating this Knowledge Graph?
Tell me about knowledge representation in relation to the episode Episode 01: Kelvin Guu, Google AI, on language models & overlooked research problems and the clip Memory Architectures